1 /* 2 * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 */ 32 #include <linux/module.h> 33 #include <linux/list.h> 34 #include <linux/workqueue.h> 35 #include <linux/skbuff.h> 36 #include <linux/timer.h> 37 #include <linux/notifier.h> 38 #include <linux/inetdevice.h> 39 #include <linux/ip.h> 40 #include <linux/tcp.h> 41 #include <linux/if_vlan.h> 42 43 #include <net/neighbour.h> 44 #include <net/netevent.h> 45 #include <net/route.h> 46 #include <net/tcp.h> 47 #include <net/ip6_route.h> 48 #include <net/addrconf.h> 49 50 #include <rdma/ib_addr.h> 51 52 #include <libcxgb_cm.h> 53 #include "iw_cxgb4.h" 54 #include "clip_tbl.h" 55 56 static char *states[] = { 57 "idle", 58 "listen", 59 "connecting", 60 "mpa_wait_req", 61 "mpa_req_sent", 62 "mpa_req_rcvd", 63 "mpa_rep_sent", 64 "fpdu_mode", 65 "aborting", 66 "closing", 67 "moribund", 68 "dead", 69 NULL, 70 }; 71 72 static int nocong; 73 module_param(nocong, int, 0644); 74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)"); 75 76 static int enable_ecn; 77 module_param(enable_ecn, int, 0644); 78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)"); 79 80 static int dack_mode = 1; 81 module_param(dack_mode, int, 0644); 82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)"); 83 84 uint c4iw_max_read_depth = 32; 85 module_param(c4iw_max_read_depth, int, 0644); 86 MODULE_PARM_DESC(c4iw_max_read_depth, 87 "Per-connection max ORD/IRD (default=32)"); 88 89 static int enable_tcp_timestamps; 90 module_param(enable_tcp_timestamps, int, 0644); 91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)"); 92 93 static int enable_tcp_sack; 94 module_param(enable_tcp_sack, int, 0644); 95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)"); 96 97 static int enable_tcp_window_scaling = 1; 98 module_param(enable_tcp_window_scaling, int, 0644); 99 MODULE_PARM_DESC(enable_tcp_window_scaling, 100 "Enable tcp window scaling (default=1)"); 101 102 int c4iw_debug; 103 module_param(c4iw_debug, int, 0644); 104 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)"); 105 106 static int peer2peer = 1; 107 module_param(peer2peer, int, 0644); 108 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)"); 109 110 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; 111 module_param(p2p_type, int, 0644); 112 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: " 113 "1=RDMA_READ 0=RDMA_WRITE (default 1)"); 114 115 static int ep_timeout_secs = 60; 116 module_param(ep_timeout_secs, int, 0644); 117 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout " 118 "in seconds (default=60)"); 119 120 static int mpa_rev = 2; 121 module_param(mpa_rev, int, 0644); 122 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, " 123 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft" 124 " compliant (default=2)"); 125 126 static int markers_enabled; 127 module_param(markers_enabled, int, 0644); 128 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)"); 129 130 static int crc_enabled = 1; 131 module_param(crc_enabled, int, 0644); 132 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)"); 133 134 static int rcv_win = 256 * 1024; 135 module_param(rcv_win, int, 0644); 136 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)"); 137 138 static int snd_win = 128 * 1024; 139 module_param(snd_win, int, 0644); 140 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)"); 141 142 static struct workqueue_struct *workq; 143 144 static struct sk_buff_head rxq; 145 146 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp); 147 static void ep_timeout(unsigned long arg); 148 static void connect_reply_upcall(struct c4iw_ep *ep, int status); 149 static int sched(struct c4iw_dev *dev, struct sk_buff *skb); 150 151 static LIST_HEAD(timeout_list); 152 static spinlock_t timeout_lock; 153 154 static void deref_cm_id(struct c4iw_ep_common *epc) 155 { 156 epc->cm_id->rem_ref(epc->cm_id); 157 epc->cm_id = NULL; 158 set_bit(CM_ID_DEREFED, &epc->history); 159 } 160 161 static void ref_cm_id(struct c4iw_ep_common *epc) 162 { 163 set_bit(CM_ID_REFED, &epc->history); 164 epc->cm_id->add_ref(epc->cm_id); 165 } 166 167 static void deref_qp(struct c4iw_ep *ep) 168 { 169 c4iw_qp_rem_ref(&ep->com.qp->ibqp); 170 clear_bit(QP_REFERENCED, &ep->com.flags); 171 set_bit(QP_DEREFED, &ep->com.history); 172 } 173 174 static void ref_qp(struct c4iw_ep *ep) 175 { 176 set_bit(QP_REFERENCED, &ep->com.flags); 177 set_bit(QP_REFED, &ep->com.history); 178 c4iw_qp_add_ref(&ep->com.qp->ibqp); 179 } 180 181 static void start_ep_timer(struct c4iw_ep *ep) 182 { 183 PDBG("%s ep %p\n", __func__, ep); 184 if (timer_pending(&ep->timer)) { 185 pr_err("%s timer already started! ep %p\n", 186 __func__, ep); 187 return; 188 } 189 clear_bit(TIMEOUT, &ep->com.flags); 190 c4iw_get_ep(&ep->com); 191 ep->timer.expires = jiffies + ep_timeout_secs * HZ; 192 ep->timer.data = (unsigned long)ep; 193 ep->timer.function = ep_timeout; 194 add_timer(&ep->timer); 195 } 196 197 static int stop_ep_timer(struct c4iw_ep *ep) 198 { 199 PDBG("%s ep %p stopping\n", __func__, ep); 200 del_timer_sync(&ep->timer); 201 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 202 c4iw_put_ep(&ep->com); 203 return 0; 204 } 205 return 1; 206 } 207 208 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb, 209 struct l2t_entry *l2e) 210 { 211 int error = 0; 212 213 if (c4iw_fatal_error(rdev)) { 214 kfree_skb(skb); 215 PDBG("%s - device in error state - dropping\n", __func__); 216 return -EIO; 217 } 218 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e); 219 if (error < 0) 220 kfree_skb(skb); 221 else if (error == NET_XMIT_DROP) 222 return -ENOMEM; 223 return error < 0 ? error : 0; 224 } 225 226 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb) 227 { 228 int error = 0; 229 230 if (c4iw_fatal_error(rdev)) { 231 kfree_skb(skb); 232 PDBG("%s - device in error state - dropping\n", __func__); 233 return -EIO; 234 } 235 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb); 236 if (error < 0) 237 kfree_skb(skb); 238 return error < 0 ? error : 0; 239 } 240 241 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb) 242 { 243 u32 len = roundup(sizeof(struct cpl_tid_release), 16); 244 245 skb = get_skb(skb, len, GFP_KERNEL); 246 if (!skb) 247 return; 248 249 cxgb_mk_tid_release(skb, len, hwtid, 0); 250 c4iw_ofld_send(rdev, skb); 251 return; 252 } 253 254 static void set_emss(struct c4iw_ep *ep, u16 opt) 255 { 256 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] - 257 ((AF_INET == ep->com.remote_addr.ss_family) ? 258 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) - 259 sizeof(struct tcphdr); 260 ep->mss = ep->emss; 261 if (TCPOPT_TSTAMP_G(opt)) 262 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4); 263 if (ep->emss < 128) 264 ep->emss = 128; 265 if (ep->emss & 7) 266 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n", 267 TCPOPT_MSS_G(opt), ep->mss, ep->emss); 268 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt), 269 ep->mss, ep->emss); 270 } 271 272 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc) 273 { 274 enum c4iw_ep_state state; 275 276 mutex_lock(&epc->mutex); 277 state = epc->state; 278 mutex_unlock(&epc->mutex); 279 return state; 280 } 281 282 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) 283 { 284 epc->state = new; 285 } 286 287 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) 288 { 289 mutex_lock(&epc->mutex); 290 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]); 291 __state_set(epc, new); 292 mutex_unlock(&epc->mutex); 293 return; 294 } 295 296 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size) 297 { 298 struct sk_buff *skb; 299 unsigned int i; 300 size_t len; 301 302 len = roundup(sizeof(union cpl_wr_size), 16); 303 for (i = 0; i < size; i++) { 304 skb = alloc_skb(len, GFP_KERNEL); 305 if (!skb) 306 goto fail; 307 skb_queue_tail(ep_skb_list, skb); 308 } 309 return 0; 310 fail: 311 skb_queue_purge(ep_skb_list); 312 return -ENOMEM; 313 } 314 315 static void *alloc_ep(int size, gfp_t gfp) 316 { 317 struct c4iw_ep_common *epc; 318 319 epc = kzalloc(size, gfp); 320 if (epc) { 321 kref_init(&epc->kref); 322 mutex_init(&epc->mutex); 323 c4iw_init_wr_wait(&epc->wr_wait); 324 } 325 PDBG("%s alloc ep %p\n", __func__, epc); 326 return epc; 327 } 328 329 static void remove_ep_tid(struct c4iw_ep *ep) 330 { 331 unsigned long flags; 332 333 spin_lock_irqsave(&ep->com.dev->lock, flags); 334 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0); 335 if (idr_is_empty(&ep->com.dev->hwtid_idr)) 336 wake_up(&ep->com.dev->wait); 337 spin_unlock_irqrestore(&ep->com.dev->lock, flags); 338 } 339 340 static void insert_ep_tid(struct c4iw_ep *ep) 341 { 342 unsigned long flags; 343 344 spin_lock_irqsave(&ep->com.dev->lock, flags); 345 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0); 346 spin_unlock_irqrestore(&ep->com.dev->lock, flags); 347 } 348 349 /* 350 * Atomically lookup the ep ptr given the tid and grab a reference on the ep. 351 */ 352 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid) 353 { 354 struct c4iw_ep *ep; 355 unsigned long flags; 356 357 spin_lock_irqsave(&dev->lock, flags); 358 ep = idr_find(&dev->hwtid_idr, tid); 359 if (ep) 360 c4iw_get_ep(&ep->com); 361 spin_unlock_irqrestore(&dev->lock, flags); 362 return ep; 363 } 364 365 /* 366 * Atomically lookup the ep ptr given the stid and grab a reference on the ep. 367 */ 368 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev, 369 unsigned int stid) 370 { 371 struct c4iw_listen_ep *ep; 372 unsigned long flags; 373 374 spin_lock_irqsave(&dev->lock, flags); 375 ep = idr_find(&dev->stid_idr, stid); 376 if (ep) 377 c4iw_get_ep(&ep->com); 378 spin_unlock_irqrestore(&dev->lock, flags); 379 return ep; 380 } 381 382 void _c4iw_free_ep(struct kref *kref) 383 { 384 struct c4iw_ep *ep; 385 386 ep = container_of(kref, struct c4iw_ep, com.kref); 387 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]); 388 if (test_bit(QP_REFERENCED, &ep->com.flags)) 389 deref_qp(ep); 390 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) { 391 if (ep->com.remote_addr.ss_family == AF_INET6) { 392 struct sockaddr_in6 *sin6 = 393 (struct sockaddr_in6 *) 394 &ep->com.local_addr; 395 396 cxgb4_clip_release( 397 ep->com.dev->rdev.lldi.ports[0], 398 (const u32 *)&sin6->sin6_addr.s6_addr, 399 1); 400 } 401 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid); 402 dst_release(ep->dst); 403 cxgb4_l2t_release(ep->l2t); 404 if (ep->mpa_skb) 405 kfree_skb(ep->mpa_skb); 406 } 407 if (!skb_queue_empty(&ep->com.ep_skb_list)) 408 skb_queue_purge(&ep->com.ep_skb_list); 409 kfree(ep); 410 } 411 412 static void release_ep_resources(struct c4iw_ep *ep) 413 { 414 set_bit(RELEASE_RESOURCES, &ep->com.flags); 415 416 /* 417 * If we have a hwtid, then remove it from the idr table 418 * so lookups will no longer find this endpoint. Otherwise 419 * we have a race where one thread finds the ep ptr just 420 * before the other thread is freeing the ep memory. 421 */ 422 if (ep->hwtid != -1) 423 remove_ep_tid(ep); 424 c4iw_put_ep(&ep->com); 425 } 426 427 static int status2errno(int status) 428 { 429 switch (status) { 430 case CPL_ERR_NONE: 431 return 0; 432 case CPL_ERR_CONN_RESET: 433 return -ECONNRESET; 434 case CPL_ERR_ARP_MISS: 435 return -EHOSTUNREACH; 436 case CPL_ERR_CONN_TIMEDOUT: 437 return -ETIMEDOUT; 438 case CPL_ERR_TCAM_FULL: 439 return -ENOMEM; 440 case CPL_ERR_CONN_EXIST: 441 return -EADDRINUSE; 442 default: 443 return -EIO; 444 } 445 } 446 447 /* 448 * Try and reuse skbs already allocated... 449 */ 450 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp) 451 { 452 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) { 453 skb_trim(skb, 0); 454 skb_get(skb); 455 skb_reset_transport_header(skb); 456 } else { 457 skb = alloc_skb(len, gfp); 458 } 459 t4_set_arp_err_handler(skb, NULL, NULL); 460 return skb; 461 } 462 463 static struct net_device *get_real_dev(struct net_device *egress_dev) 464 { 465 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev; 466 } 467 468 static void arp_failure_discard(void *handle, struct sk_buff *skb) 469 { 470 pr_err(MOD "ARP failure\n"); 471 kfree_skb(skb); 472 } 473 474 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb) 475 { 476 pr_err("ARP failure during MPA Negotiation - Closing Connection\n"); 477 } 478 479 enum { 480 NUM_FAKE_CPLS = 2, 481 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0, 482 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1, 483 }; 484 485 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb) 486 { 487 struct c4iw_ep *ep; 488 489 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))); 490 release_ep_resources(ep); 491 return 0; 492 } 493 494 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb) 495 { 496 struct c4iw_ep *ep; 497 498 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))); 499 c4iw_put_ep(&ep->parent_ep->com); 500 release_ep_resources(ep); 501 return 0; 502 } 503 504 /* 505 * Fake up a special CPL opcode and call sched() so process_work() will call 506 * _put_ep_safe() in a safe context to free the ep resources. This is needed 507 * because ARP error handlers are called in an ATOMIC context, and 508 * _c4iw_free_ep() needs to block. 509 */ 510 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb, 511 int cpl) 512 { 513 struct cpl_act_establish *rpl = cplhdr(skb); 514 515 /* Set our special ARP_FAILURE opcode */ 516 rpl->ot.opcode = cpl; 517 518 /* 519 * Save ep in the skb->cb area, after where sched() will save the dev 520 * ptr. 521 */ 522 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep; 523 sched(ep->com.dev, skb); 524 } 525 526 /* Handle an ARP failure for an accept */ 527 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb) 528 { 529 struct c4iw_ep *ep = handle; 530 531 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n", 532 ep->hwtid); 533 534 __state_set(&ep->com, DEAD); 535 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE); 536 } 537 538 /* 539 * Handle an ARP failure for an active open. 540 */ 541 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb) 542 { 543 struct c4iw_ep *ep = handle; 544 545 printk(KERN_ERR MOD "ARP failure during connect\n"); 546 connect_reply_upcall(ep, -EHOSTUNREACH); 547 __state_set(&ep->com, DEAD); 548 if (ep->com.remote_addr.ss_family == AF_INET6) { 549 struct sockaddr_in6 *sin6 = 550 (struct sockaddr_in6 *)&ep->com.local_addr; 551 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 552 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 553 } 554 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 555 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 556 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE); 557 } 558 559 /* 560 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant 561 * and send it along. 562 */ 563 static void abort_arp_failure(void *handle, struct sk_buff *skb) 564 { 565 int ret; 566 struct c4iw_ep *ep = handle; 567 struct c4iw_rdev *rdev = &ep->com.dev->rdev; 568 struct cpl_abort_req *req = cplhdr(skb); 569 570 PDBG("%s rdev %p\n", __func__, rdev); 571 req->cmd = CPL_ABORT_NO_RST; 572 ret = c4iw_ofld_send(rdev, skb); 573 if (ret) { 574 __state_set(&ep->com, DEAD); 575 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE); 576 } 577 } 578 579 static int send_flowc(struct c4iw_ep *ep) 580 { 581 struct fw_flowc_wr *flowc; 582 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list); 583 int i; 584 u16 vlan = ep->l2t->vlan; 585 int nparams; 586 587 if (WARN_ON(!skb)) 588 return -ENOMEM; 589 590 if (vlan == CPL_L2T_VLAN_NONE) 591 nparams = 8; 592 else 593 nparams = 9; 594 595 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN); 596 597 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) | 598 FW_FLOWC_WR_NPARAMS_V(nparams)); 599 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN, 600 16)) | FW_WR_FLOWID_V(ep->hwtid)); 601 602 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN; 603 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V 604 (ep->com.dev->rdev.lldi.pf)); 605 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH; 606 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan); 607 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT; 608 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan); 609 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID; 610 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid); 611 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT; 612 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq); 613 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT; 614 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq); 615 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF; 616 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win); 617 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS; 618 flowc->mnemval[7].val = cpu_to_be32(ep->emss); 619 if (nparams == 9) { 620 u16 pri; 621 622 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT; 623 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS; 624 flowc->mnemval[8].val = cpu_to_be32(pri); 625 } else { 626 /* Pad WR to 16 byte boundary */ 627 flowc->mnemval[8].mnemonic = 0; 628 flowc->mnemval[8].val = 0; 629 } 630 for (i = 0; i < 9; i++) { 631 flowc->mnemval[i].r4[0] = 0; 632 flowc->mnemval[i].r4[1] = 0; 633 flowc->mnemval[i].r4[2] = 0; 634 } 635 636 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 637 return c4iw_ofld_send(&ep->com.dev->rdev, skb); 638 } 639 640 static int send_halfclose(struct c4iw_ep *ep) 641 { 642 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list); 643 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16); 644 645 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 646 if (WARN_ON(!skb)) 647 return -ENOMEM; 648 649 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx, 650 NULL, arp_failure_discard); 651 652 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 653 } 654 655 static int send_abort(struct c4iw_ep *ep) 656 { 657 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16); 658 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list); 659 660 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 661 if (WARN_ON(!req_skb)) 662 return -ENOMEM; 663 664 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx, 665 ep, abort_arp_failure); 666 667 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t); 668 } 669 670 static int send_connect(struct c4iw_ep *ep) 671 { 672 struct cpl_act_open_req *req = NULL; 673 struct cpl_t5_act_open_req *t5req = NULL; 674 struct cpl_t6_act_open_req *t6req = NULL; 675 struct cpl_act_open_req6 *req6 = NULL; 676 struct cpl_t5_act_open_req6 *t5req6 = NULL; 677 struct cpl_t6_act_open_req6 *t6req6 = NULL; 678 struct sk_buff *skb; 679 u64 opt0; 680 u32 opt2; 681 unsigned int mtu_idx; 682 u32 wscale; 683 int win, sizev4, sizev6, wrlen; 684 struct sockaddr_in *la = (struct sockaddr_in *) 685 &ep->com.local_addr; 686 struct sockaddr_in *ra = (struct sockaddr_in *) 687 &ep->com.remote_addr; 688 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *) 689 &ep->com.local_addr; 690 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *) 691 &ep->com.remote_addr; 692 int ret; 693 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type; 694 u32 isn = (prandom_u32() & ~7UL) - 1; 695 696 switch (CHELSIO_CHIP_VERSION(adapter_type)) { 697 case CHELSIO_T4: 698 sizev4 = sizeof(struct cpl_act_open_req); 699 sizev6 = sizeof(struct cpl_act_open_req6); 700 break; 701 case CHELSIO_T5: 702 sizev4 = sizeof(struct cpl_t5_act_open_req); 703 sizev6 = sizeof(struct cpl_t5_act_open_req6); 704 break; 705 case CHELSIO_T6: 706 sizev4 = sizeof(struct cpl_t6_act_open_req); 707 sizev6 = sizeof(struct cpl_t6_act_open_req6); 708 break; 709 default: 710 pr_err("T%d Chip is not supported\n", 711 CHELSIO_CHIP_VERSION(adapter_type)); 712 return -EINVAL; 713 } 714 715 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ? 716 roundup(sizev4, 16) : 717 roundup(sizev6, 16); 718 719 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid); 720 721 skb = get_skb(NULL, wrlen, GFP_KERNEL); 722 if (!skb) { 723 printk(KERN_ERR MOD "%s - failed to alloc skb.\n", 724 __func__); 725 return -ENOMEM; 726 } 727 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx); 728 729 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 730 enable_tcp_timestamps, 731 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1); 732 wscale = cxgb_compute_wscale(rcv_win); 733 734 /* 735 * Specify the largest window that will fit in opt0. The 736 * remainder will be specified in the rx_data_ack. 737 */ 738 win = ep->rcv_win >> 10; 739 if (win > RCV_BUFSIZ_M) 740 win = RCV_BUFSIZ_M; 741 742 opt0 = (nocong ? NO_CONG_F : 0) | 743 KEEP_ALIVE_F | 744 DELACK_F | 745 WND_SCALE_V(wscale) | 746 MSS_IDX_V(mtu_idx) | 747 L2T_IDX_V(ep->l2t->idx) | 748 TX_CHAN_V(ep->tx_chan) | 749 SMAC_SEL_V(ep->smac_idx) | 750 DSCP_V(ep->tos >> 2) | 751 ULP_MODE_V(ULP_MODE_TCPDDP) | 752 RCV_BUFSIZ_V(win); 753 opt2 = RX_CHANNEL_V(0) | 754 CCTRL_ECN_V(enable_ecn) | 755 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid); 756 if (enable_tcp_timestamps) 757 opt2 |= TSTAMPS_EN_F; 758 if (enable_tcp_sack) 759 opt2 |= SACK_EN_F; 760 if (wscale && enable_tcp_window_scaling) 761 opt2 |= WND_SCALE_EN_F; 762 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) { 763 if (peer2peer) 764 isn += 4; 765 766 opt2 |= T5_OPT_2_VALID_F; 767 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE); 768 opt2 |= T5_ISS_F; 769 } 770 771 if (ep->com.remote_addr.ss_family == AF_INET6) 772 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0], 773 (const u32 *)&la6->sin6_addr.s6_addr, 1); 774 775 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure); 776 777 if (ep->com.remote_addr.ss_family == AF_INET) { 778 switch (CHELSIO_CHIP_VERSION(adapter_type)) { 779 case CHELSIO_T4: 780 req = (struct cpl_act_open_req *)skb_put(skb, wrlen); 781 INIT_TP_WR(req, 0); 782 break; 783 case CHELSIO_T5: 784 t5req = (struct cpl_t5_act_open_req *)skb_put(skb, 785 wrlen); 786 INIT_TP_WR(t5req, 0); 787 req = (struct cpl_act_open_req *)t5req; 788 break; 789 case CHELSIO_T6: 790 t6req = (struct cpl_t6_act_open_req *)skb_put(skb, 791 wrlen); 792 INIT_TP_WR(t6req, 0); 793 req = (struct cpl_act_open_req *)t6req; 794 t5req = (struct cpl_t5_act_open_req *)t6req; 795 break; 796 default: 797 pr_err("T%d Chip is not supported\n", 798 CHELSIO_CHIP_VERSION(adapter_type)); 799 ret = -EINVAL; 800 goto clip_release; 801 } 802 803 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 804 ((ep->rss_qid<<14) | ep->atid))); 805 req->local_port = la->sin_port; 806 req->peer_port = ra->sin_port; 807 req->local_ip = la->sin_addr.s_addr; 808 req->peer_ip = ra->sin_addr.s_addr; 809 req->opt0 = cpu_to_be64(opt0); 810 811 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) { 812 req->params = cpu_to_be32(cxgb4_select_ntuple( 813 ep->com.dev->rdev.lldi.ports[0], 814 ep->l2t)); 815 req->opt2 = cpu_to_be32(opt2); 816 } else { 817 t5req->params = cpu_to_be64(FILTER_TUPLE_V( 818 cxgb4_select_ntuple( 819 ep->com.dev->rdev.lldi.ports[0], 820 ep->l2t))); 821 t5req->rsvd = cpu_to_be32(isn); 822 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd); 823 t5req->opt2 = cpu_to_be32(opt2); 824 } 825 } else { 826 switch (CHELSIO_CHIP_VERSION(adapter_type)) { 827 case CHELSIO_T4: 828 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen); 829 INIT_TP_WR(req6, 0); 830 break; 831 case CHELSIO_T5: 832 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb, 833 wrlen); 834 INIT_TP_WR(t5req6, 0); 835 req6 = (struct cpl_act_open_req6 *)t5req6; 836 break; 837 case CHELSIO_T6: 838 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb, 839 wrlen); 840 INIT_TP_WR(t6req6, 0); 841 req6 = (struct cpl_act_open_req6 *)t6req6; 842 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6; 843 break; 844 default: 845 pr_err("T%d Chip is not supported\n", 846 CHELSIO_CHIP_VERSION(adapter_type)); 847 ret = -EINVAL; 848 goto clip_release; 849 } 850 851 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 852 ((ep->rss_qid<<14)|ep->atid))); 853 req6->local_port = la6->sin6_port; 854 req6->peer_port = ra6->sin6_port; 855 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr)); 856 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8)); 857 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr)); 858 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8)); 859 req6->opt0 = cpu_to_be64(opt0); 860 861 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) { 862 req6->params = cpu_to_be32(cxgb4_select_ntuple( 863 ep->com.dev->rdev.lldi.ports[0], 864 ep->l2t)); 865 req6->opt2 = cpu_to_be32(opt2); 866 } else { 867 t5req6->params = cpu_to_be64(FILTER_TUPLE_V( 868 cxgb4_select_ntuple( 869 ep->com.dev->rdev.lldi.ports[0], 870 ep->l2t))); 871 t5req6->rsvd = cpu_to_be32(isn); 872 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd); 873 t5req6->opt2 = cpu_to_be32(opt2); 874 } 875 } 876 877 set_bit(ACT_OPEN_REQ, &ep->com.history); 878 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 879 clip_release: 880 if (ret && ep->com.remote_addr.ss_family == AF_INET6) 881 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 882 (const u32 *)&la6->sin6_addr.s6_addr, 1); 883 return ret; 884 } 885 886 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb, 887 u8 mpa_rev_to_use) 888 { 889 int mpalen, wrlen, ret; 890 struct fw_ofld_tx_data_wr *req; 891 struct mpa_message *mpa; 892 struct mpa_v2_conn_params mpa_v2_params; 893 894 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen); 895 896 BUG_ON(skb_cloned(skb)); 897 898 mpalen = sizeof(*mpa) + ep->plen; 899 if (mpa_rev_to_use == 2) 900 mpalen += sizeof(struct mpa_v2_conn_params); 901 wrlen = roundup(mpalen + sizeof *req, 16); 902 skb = get_skb(skb, wrlen, GFP_KERNEL); 903 if (!skb) { 904 connect_reply_upcall(ep, -ENOMEM); 905 return -ENOMEM; 906 } 907 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 908 909 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen); 910 memset(req, 0, wrlen); 911 req->op_to_immdlen = cpu_to_be32( 912 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 913 FW_WR_COMPL_F | 914 FW_WR_IMMDLEN_V(mpalen)); 915 req->flowid_len16 = cpu_to_be32( 916 FW_WR_FLOWID_V(ep->hwtid) | 917 FW_WR_LEN16_V(wrlen >> 4)); 918 req->plen = cpu_to_be32(mpalen); 919 req->tunnel_to_proxy = cpu_to_be32( 920 FW_OFLD_TX_DATA_WR_FLUSH_F | 921 FW_OFLD_TX_DATA_WR_SHOVE_F); 922 923 mpa = (struct mpa_message *)(req + 1); 924 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); 925 926 mpa->flags = 0; 927 if (crc_enabled) 928 mpa->flags |= MPA_CRC; 929 if (markers_enabled) { 930 mpa->flags |= MPA_MARKERS; 931 ep->mpa_attr.recv_marker_enabled = 1; 932 } else { 933 ep->mpa_attr.recv_marker_enabled = 0; 934 } 935 if (mpa_rev_to_use == 2) 936 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 937 938 mpa->private_data_size = htons(ep->plen); 939 mpa->revision = mpa_rev_to_use; 940 if (mpa_rev_to_use == 1) { 941 ep->tried_with_mpa_v1 = 1; 942 ep->retry_with_mpa_v1 = 0; 943 } 944 945 if (mpa_rev_to_use == 2) { 946 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 947 sizeof (struct mpa_v2_conn_params)); 948 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird, 949 ep->ord); 950 mpa_v2_params.ird = htons((u16)ep->ird); 951 mpa_v2_params.ord = htons((u16)ep->ord); 952 953 if (peer2peer) { 954 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 955 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) 956 mpa_v2_params.ord |= 957 htons(MPA_V2_RDMA_WRITE_RTR); 958 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) 959 mpa_v2_params.ord |= 960 htons(MPA_V2_RDMA_READ_RTR); 961 } 962 memcpy(mpa->private_data, &mpa_v2_params, 963 sizeof(struct mpa_v2_conn_params)); 964 965 if (ep->plen) 966 memcpy(mpa->private_data + 967 sizeof(struct mpa_v2_conn_params), 968 ep->mpa_pkt + sizeof(*mpa), ep->plen); 969 } else 970 if (ep->plen) 971 memcpy(mpa->private_data, 972 ep->mpa_pkt + sizeof(*mpa), ep->plen); 973 974 /* 975 * Reference the mpa skb. This ensures the data area 976 * will remain in memory until the hw acks the tx. 977 * Function fw4_ack() will deref it. 978 */ 979 skb_get(skb); 980 t4_set_arp_err_handler(skb, NULL, arp_failure_discard); 981 BUG_ON(ep->mpa_skb); 982 ep->mpa_skb = skb; 983 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 984 if (ret) 985 return ret; 986 start_ep_timer(ep); 987 __state_set(&ep->com, MPA_REQ_SENT); 988 ep->mpa_attr.initiator = 1; 989 ep->snd_seq += mpalen; 990 return ret; 991 } 992 993 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen) 994 { 995 int mpalen, wrlen; 996 struct fw_ofld_tx_data_wr *req; 997 struct mpa_message *mpa; 998 struct sk_buff *skb; 999 struct mpa_v2_conn_params mpa_v2_params; 1000 1001 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen); 1002 1003 mpalen = sizeof(*mpa) + plen; 1004 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) 1005 mpalen += sizeof(struct mpa_v2_conn_params); 1006 wrlen = roundup(mpalen + sizeof *req, 16); 1007 1008 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1009 if (!skb) { 1010 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__); 1011 return -ENOMEM; 1012 } 1013 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1014 1015 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen); 1016 memset(req, 0, wrlen); 1017 req->op_to_immdlen = cpu_to_be32( 1018 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 1019 FW_WR_COMPL_F | 1020 FW_WR_IMMDLEN_V(mpalen)); 1021 req->flowid_len16 = cpu_to_be32( 1022 FW_WR_FLOWID_V(ep->hwtid) | 1023 FW_WR_LEN16_V(wrlen >> 4)); 1024 req->plen = cpu_to_be32(mpalen); 1025 req->tunnel_to_proxy = cpu_to_be32( 1026 FW_OFLD_TX_DATA_WR_FLUSH_F | 1027 FW_OFLD_TX_DATA_WR_SHOVE_F); 1028 1029 mpa = (struct mpa_message *)(req + 1); 1030 memset(mpa, 0, sizeof(*mpa)); 1031 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1032 mpa->flags = MPA_REJECT; 1033 mpa->revision = ep->mpa_attr.version; 1034 mpa->private_data_size = htons(plen); 1035 1036 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1037 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1038 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1039 sizeof (struct mpa_v2_conn_params)); 1040 mpa_v2_params.ird = htons(((u16)ep->ird) | 1041 (peer2peer ? MPA_V2_PEER2PEER_MODEL : 1042 0)); 1043 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ? 1044 (p2p_type == 1045 FW_RI_INIT_P2PTYPE_RDMA_WRITE ? 1046 MPA_V2_RDMA_WRITE_RTR : p2p_type == 1047 FW_RI_INIT_P2PTYPE_READ_REQ ? 1048 MPA_V2_RDMA_READ_RTR : 0) : 0)); 1049 memcpy(mpa->private_data, &mpa_v2_params, 1050 sizeof(struct mpa_v2_conn_params)); 1051 1052 if (ep->plen) 1053 memcpy(mpa->private_data + 1054 sizeof(struct mpa_v2_conn_params), pdata, plen); 1055 } else 1056 if (plen) 1057 memcpy(mpa->private_data, pdata, plen); 1058 1059 /* 1060 * Reference the mpa skb again. This ensures the data area 1061 * will remain in memory until the hw acks the tx. 1062 * Function fw4_ack() will deref it. 1063 */ 1064 skb_get(skb); 1065 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1066 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure); 1067 BUG_ON(ep->mpa_skb); 1068 ep->mpa_skb = skb; 1069 ep->snd_seq += mpalen; 1070 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1071 } 1072 1073 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen) 1074 { 1075 int mpalen, wrlen; 1076 struct fw_ofld_tx_data_wr *req; 1077 struct mpa_message *mpa; 1078 struct sk_buff *skb; 1079 struct mpa_v2_conn_params mpa_v2_params; 1080 1081 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen); 1082 1083 mpalen = sizeof(*mpa) + plen; 1084 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) 1085 mpalen += sizeof(struct mpa_v2_conn_params); 1086 wrlen = roundup(mpalen + sizeof *req, 16); 1087 1088 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1089 if (!skb) { 1090 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__); 1091 return -ENOMEM; 1092 } 1093 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1094 1095 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen); 1096 memset(req, 0, wrlen); 1097 req->op_to_immdlen = cpu_to_be32( 1098 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 1099 FW_WR_COMPL_F | 1100 FW_WR_IMMDLEN_V(mpalen)); 1101 req->flowid_len16 = cpu_to_be32( 1102 FW_WR_FLOWID_V(ep->hwtid) | 1103 FW_WR_LEN16_V(wrlen >> 4)); 1104 req->plen = cpu_to_be32(mpalen); 1105 req->tunnel_to_proxy = cpu_to_be32( 1106 FW_OFLD_TX_DATA_WR_FLUSH_F | 1107 FW_OFLD_TX_DATA_WR_SHOVE_F); 1108 1109 mpa = (struct mpa_message *)(req + 1); 1110 memset(mpa, 0, sizeof(*mpa)); 1111 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1112 mpa->flags = 0; 1113 if (ep->mpa_attr.crc_enabled) 1114 mpa->flags |= MPA_CRC; 1115 if (ep->mpa_attr.recv_marker_enabled) 1116 mpa->flags |= MPA_MARKERS; 1117 mpa->revision = ep->mpa_attr.version; 1118 mpa->private_data_size = htons(plen); 1119 1120 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1121 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1122 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1123 sizeof (struct mpa_v2_conn_params)); 1124 mpa_v2_params.ird = htons((u16)ep->ird); 1125 mpa_v2_params.ord = htons((u16)ep->ord); 1126 if (peer2peer && (ep->mpa_attr.p2p_type != 1127 FW_RI_INIT_P2PTYPE_DISABLED)) { 1128 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1129 1130 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) 1131 mpa_v2_params.ord |= 1132 htons(MPA_V2_RDMA_WRITE_RTR); 1133 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) 1134 mpa_v2_params.ord |= 1135 htons(MPA_V2_RDMA_READ_RTR); 1136 } 1137 1138 memcpy(mpa->private_data, &mpa_v2_params, 1139 sizeof(struct mpa_v2_conn_params)); 1140 1141 if (ep->plen) 1142 memcpy(mpa->private_data + 1143 sizeof(struct mpa_v2_conn_params), pdata, plen); 1144 } else 1145 if (plen) 1146 memcpy(mpa->private_data, pdata, plen); 1147 1148 /* 1149 * Reference the mpa skb. This ensures the data area 1150 * will remain in memory until the hw acks the tx. 1151 * Function fw4_ack() will deref it. 1152 */ 1153 skb_get(skb); 1154 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure); 1155 ep->mpa_skb = skb; 1156 __state_set(&ep->com, MPA_REP_SENT); 1157 ep->snd_seq += mpalen; 1158 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1159 } 1160 1161 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb) 1162 { 1163 struct c4iw_ep *ep; 1164 struct cpl_act_establish *req = cplhdr(skb); 1165 unsigned int tid = GET_TID(req); 1166 unsigned int atid = TID_TID_G(ntohl(req->tos_atid)); 1167 struct tid_info *t = dev->rdev.lldi.tids; 1168 int ret; 1169 1170 ep = lookup_atid(t, atid); 1171 1172 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid, 1173 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn)); 1174 1175 mutex_lock(&ep->com.mutex); 1176 dst_confirm(ep->dst); 1177 1178 /* setup the hwtid for this connection */ 1179 ep->hwtid = tid; 1180 cxgb4_insert_tid(t, ep, tid); 1181 insert_ep_tid(ep); 1182 1183 ep->snd_seq = be32_to_cpu(req->snd_isn); 1184 ep->rcv_seq = be32_to_cpu(req->rcv_isn); 1185 1186 set_emss(ep, ntohs(req->tcp_opt)); 1187 1188 /* dealloc the atid */ 1189 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid); 1190 cxgb4_free_atid(t, atid); 1191 set_bit(ACT_ESTAB, &ep->com.history); 1192 1193 /* start MPA negotiation */ 1194 ret = send_flowc(ep); 1195 if (ret) 1196 goto err; 1197 if (ep->retry_with_mpa_v1) 1198 ret = send_mpa_req(ep, skb, 1); 1199 else 1200 ret = send_mpa_req(ep, skb, mpa_rev); 1201 if (ret) 1202 goto err; 1203 mutex_unlock(&ep->com.mutex); 1204 return 0; 1205 err: 1206 mutex_unlock(&ep->com.mutex); 1207 connect_reply_upcall(ep, -ENOMEM); 1208 c4iw_ep_disconnect(ep, 0, GFP_KERNEL); 1209 return 0; 1210 } 1211 1212 static void close_complete_upcall(struct c4iw_ep *ep, int status) 1213 { 1214 struct iw_cm_event event; 1215 1216 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1217 memset(&event, 0, sizeof(event)); 1218 event.event = IW_CM_EVENT_CLOSE; 1219 event.status = status; 1220 if (ep->com.cm_id) { 1221 PDBG("close complete delivered ep %p cm_id %p tid %u\n", 1222 ep, ep->com.cm_id, ep->hwtid); 1223 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1224 deref_cm_id(&ep->com); 1225 set_bit(CLOSE_UPCALL, &ep->com.history); 1226 } 1227 } 1228 1229 static void peer_close_upcall(struct c4iw_ep *ep) 1230 { 1231 struct iw_cm_event event; 1232 1233 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1234 memset(&event, 0, sizeof(event)); 1235 event.event = IW_CM_EVENT_DISCONNECT; 1236 if (ep->com.cm_id) { 1237 PDBG("peer close delivered ep %p cm_id %p tid %u\n", 1238 ep, ep->com.cm_id, ep->hwtid); 1239 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1240 set_bit(DISCONN_UPCALL, &ep->com.history); 1241 } 1242 } 1243 1244 static void peer_abort_upcall(struct c4iw_ep *ep) 1245 { 1246 struct iw_cm_event event; 1247 1248 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1249 memset(&event, 0, sizeof(event)); 1250 event.event = IW_CM_EVENT_CLOSE; 1251 event.status = -ECONNRESET; 1252 if (ep->com.cm_id) { 1253 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep, 1254 ep->com.cm_id, ep->hwtid); 1255 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1256 deref_cm_id(&ep->com); 1257 set_bit(ABORT_UPCALL, &ep->com.history); 1258 } 1259 } 1260 1261 static void connect_reply_upcall(struct c4iw_ep *ep, int status) 1262 { 1263 struct iw_cm_event event; 1264 1265 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status); 1266 memset(&event, 0, sizeof(event)); 1267 event.event = IW_CM_EVENT_CONNECT_REPLY; 1268 event.status = status; 1269 memcpy(&event.local_addr, &ep->com.local_addr, 1270 sizeof(ep->com.local_addr)); 1271 memcpy(&event.remote_addr, &ep->com.remote_addr, 1272 sizeof(ep->com.remote_addr)); 1273 1274 if ((status == 0) || (status == -ECONNREFUSED)) { 1275 if (!ep->tried_with_mpa_v1) { 1276 /* this means MPA_v2 is used */ 1277 event.ord = ep->ird; 1278 event.ird = ep->ord; 1279 event.private_data_len = ep->plen - 1280 sizeof(struct mpa_v2_conn_params); 1281 event.private_data = ep->mpa_pkt + 1282 sizeof(struct mpa_message) + 1283 sizeof(struct mpa_v2_conn_params); 1284 } else { 1285 /* this means MPA_v1 is used */ 1286 event.ord = cur_max_read_depth(ep->com.dev); 1287 event.ird = cur_max_read_depth(ep->com.dev); 1288 event.private_data_len = ep->plen; 1289 event.private_data = ep->mpa_pkt + 1290 sizeof(struct mpa_message); 1291 } 1292 } 1293 1294 PDBG("%s ep %p tid %u status %d\n", __func__, ep, 1295 ep->hwtid, status); 1296 set_bit(CONN_RPL_UPCALL, &ep->com.history); 1297 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1298 1299 if (status < 0) 1300 deref_cm_id(&ep->com); 1301 } 1302 1303 static int connect_request_upcall(struct c4iw_ep *ep) 1304 { 1305 struct iw_cm_event event; 1306 int ret; 1307 1308 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1309 memset(&event, 0, sizeof(event)); 1310 event.event = IW_CM_EVENT_CONNECT_REQUEST; 1311 memcpy(&event.local_addr, &ep->com.local_addr, 1312 sizeof(ep->com.local_addr)); 1313 memcpy(&event.remote_addr, &ep->com.remote_addr, 1314 sizeof(ep->com.remote_addr)); 1315 event.provider_data = ep; 1316 if (!ep->tried_with_mpa_v1) { 1317 /* this means MPA_v2 is used */ 1318 event.ord = ep->ord; 1319 event.ird = ep->ird; 1320 event.private_data_len = ep->plen - 1321 sizeof(struct mpa_v2_conn_params); 1322 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + 1323 sizeof(struct mpa_v2_conn_params); 1324 } else { 1325 /* this means MPA_v1 is used. Send max supported */ 1326 event.ord = cur_max_read_depth(ep->com.dev); 1327 event.ird = cur_max_read_depth(ep->com.dev); 1328 event.private_data_len = ep->plen; 1329 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 1330 } 1331 c4iw_get_ep(&ep->com); 1332 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, 1333 &event); 1334 if (ret) 1335 c4iw_put_ep(&ep->com); 1336 set_bit(CONNREQ_UPCALL, &ep->com.history); 1337 c4iw_put_ep(&ep->parent_ep->com); 1338 return ret; 1339 } 1340 1341 static void established_upcall(struct c4iw_ep *ep) 1342 { 1343 struct iw_cm_event event; 1344 1345 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1346 memset(&event, 0, sizeof(event)); 1347 event.event = IW_CM_EVENT_ESTABLISHED; 1348 event.ird = ep->ord; 1349 event.ord = ep->ird; 1350 if (ep->com.cm_id) { 1351 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1352 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1353 set_bit(ESTAB_UPCALL, &ep->com.history); 1354 } 1355 } 1356 1357 static int update_rx_credits(struct c4iw_ep *ep, u32 credits) 1358 { 1359 struct sk_buff *skb; 1360 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16); 1361 u32 credit_dack; 1362 1363 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits); 1364 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1365 if (!skb) { 1366 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n"); 1367 return 0; 1368 } 1369 1370 /* 1371 * If we couldn't specify the entire rcv window at connection setup 1372 * due to the limit in the number of bits in the RCV_BUFSIZ field, 1373 * then add the overage in to the credits returned. 1374 */ 1375 if (ep->rcv_win > RCV_BUFSIZ_M * 1024) 1376 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024; 1377 1378 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F | 1379 RX_DACK_MODE_V(dack_mode); 1380 1381 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx, 1382 credit_dack); 1383 1384 c4iw_ofld_send(&ep->com.dev->rdev, skb); 1385 return credits; 1386 } 1387 1388 #define RELAXED_IRD_NEGOTIATION 1 1389 1390 /* 1391 * process_mpa_reply - process streaming mode MPA reply 1392 * 1393 * Returns: 1394 * 1395 * 0 upon success indicating a connect request was delivered to the ULP 1396 * or the mpa request is incomplete but valid so far. 1397 * 1398 * 1 if a failure requires the caller to close the connection. 1399 * 1400 * 2 if a failure requires the caller to abort the connection. 1401 */ 1402 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb) 1403 { 1404 struct mpa_message *mpa; 1405 struct mpa_v2_conn_params *mpa_v2_params; 1406 u16 plen; 1407 u16 resp_ird, resp_ord; 1408 u8 rtr_mismatch = 0, insuff_ird = 0; 1409 struct c4iw_qp_attributes attrs; 1410 enum c4iw_qp_attr_mask mask; 1411 int err; 1412 int disconnect = 0; 1413 1414 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1415 1416 /* 1417 * If we get more than the supported amount of private data 1418 * then we must fail this connection. 1419 */ 1420 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) { 1421 err = -EINVAL; 1422 goto err_stop_timer; 1423 } 1424 1425 /* 1426 * copy the new data into our accumulation buffer. 1427 */ 1428 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]), 1429 skb->len); 1430 ep->mpa_pkt_len += skb->len; 1431 1432 /* 1433 * if we don't even have the mpa message, then bail. 1434 */ 1435 if (ep->mpa_pkt_len < sizeof(*mpa)) 1436 return 0; 1437 mpa = (struct mpa_message *) ep->mpa_pkt; 1438 1439 /* Validate MPA header. */ 1440 if (mpa->revision > mpa_rev) { 1441 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d," 1442 " Received = %d\n", __func__, mpa_rev, mpa->revision); 1443 err = -EPROTO; 1444 goto err_stop_timer; 1445 } 1446 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { 1447 err = -EPROTO; 1448 goto err_stop_timer; 1449 } 1450 1451 plen = ntohs(mpa->private_data_size); 1452 1453 /* 1454 * Fail if there's too much private data. 1455 */ 1456 if (plen > MPA_MAX_PRIVATE_DATA) { 1457 err = -EPROTO; 1458 goto err_stop_timer; 1459 } 1460 1461 /* 1462 * If plen does not account for pkt size 1463 */ 1464 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1465 err = -EPROTO; 1466 goto err_stop_timer; 1467 } 1468 1469 ep->plen = (u8) plen; 1470 1471 /* 1472 * If we don't have all the pdata yet, then bail. 1473 * We'll continue process when more data arrives. 1474 */ 1475 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 1476 return 0; 1477 1478 if (mpa->flags & MPA_REJECT) { 1479 err = -ECONNREFUSED; 1480 goto err_stop_timer; 1481 } 1482 1483 /* 1484 * Stop mpa timer. If it expired, then 1485 * we ignore the MPA reply. process_timeout() 1486 * will abort the connection. 1487 */ 1488 if (stop_ep_timer(ep)) 1489 return 0; 1490 1491 /* 1492 * If we get here we have accumulated the entire mpa 1493 * start reply message including private data. And 1494 * the MPA header is valid. 1495 */ 1496 __state_set(&ep->com, FPDU_MODE); 1497 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1498 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1499 ep->mpa_attr.version = mpa->revision; 1500 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1501 1502 if (mpa->revision == 2) { 1503 ep->mpa_attr.enhanced_rdma_conn = 1504 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1505 if (ep->mpa_attr.enhanced_rdma_conn) { 1506 mpa_v2_params = (struct mpa_v2_conn_params *) 1507 (ep->mpa_pkt + sizeof(*mpa)); 1508 resp_ird = ntohs(mpa_v2_params->ird) & 1509 MPA_V2_IRD_ORD_MASK; 1510 resp_ord = ntohs(mpa_v2_params->ord) & 1511 MPA_V2_IRD_ORD_MASK; 1512 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n", 1513 __func__, resp_ird, resp_ord, ep->ird, ep->ord); 1514 1515 /* 1516 * This is a double-check. Ideally, below checks are 1517 * not required since ird/ord stuff has been taken 1518 * care of in c4iw_accept_cr 1519 */ 1520 if (ep->ird < resp_ord) { 1521 if (RELAXED_IRD_NEGOTIATION && resp_ord <= 1522 ep->com.dev->rdev.lldi.max_ordird_qp) 1523 ep->ird = resp_ord; 1524 else 1525 insuff_ird = 1; 1526 } else if (ep->ird > resp_ord) { 1527 ep->ird = resp_ord; 1528 } 1529 if (ep->ord > resp_ird) { 1530 if (RELAXED_IRD_NEGOTIATION) 1531 ep->ord = resp_ird; 1532 else 1533 insuff_ird = 1; 1534 } 1535 if (insuff_ird) { 1536 err = -ENOMEM; 1537 ep->ird = resp_ord; 1538 ep->ord = resp_ird; 1539 } 1540 1541 if (ntohs(mpa_v2_params->ird) & 1542 MPA_V2_PEER2PEER_MODEL) { 1543 if (ntohs(mpa_v2_params->ord) & 1544 MPA_V2_RDMA_WRITE_RTR) 1545 ep->mpa_attr.p2p_type = 1546 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 1547 else if (ntohs(mpa_v2_params->ord) & 1548 MPA_V2_RDMA_READ_RTR) 1549 ep->mpa_attr.p2p_type = 1550 FW_RI_INIT_P2PTYPE_READ_REQ; 1551 } 1552 } 1553 } else if (mpa->revision == 1) 1554 if (peer2peer) 1555 ep->mpa_attr.p2p_type = p2p_type; 1556 1557 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, " 1558 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = " 1559 "%d\n", __func__, ep->mpa_attr.crc_enabled, 1560 ep->mpa_attr.recv_marker_enabled, 1561 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 1562 ep->mpa_attr.p2p_type, p2p_type); 1563 1564 /* 1565 * If responder's RTR does not match with that of initiator, assign 1566 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not 1567 * generated when moving QP to RTS state. 1568 * A TERM message will be sent after QP has moved to RTS state 1569 */ 1570 if ((ep->mpa_attr.version == 2) && peer2peer && 1571 (ep->mpa_attr.p2p_type != p2p_type)) { 1572 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1573 rtr_mismatch = 1; 1574 } 1575 1576 attrs.mpa_attr = ep->mpa_attr; 1577 attrs.max_ird = ep->ird; 1578 attrs.max_ord = ep->ord; 1579 attrs.llp_stream_handle = ep; 1580 attrs.next_state = C4IW_QP_STATE_RTS; 1581 1582 mask = C4IW_QP_ATTR_NEXT_STATE | 1583 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | 1584 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; 1585 1586 /* bind QP and TID with INIT_WR */ 1587 err = c4iw_modify_qp(ep->com.qp->rhp, 1588 ep->com.qp, mask, &attrs, 1); 1589 if (err) 1590 goto err; 1591 1592 /* 1593 * If responder's RTR requirement did not match with what initiator 1594 * supports, generate TERM message 1595 */ 1596 if (rtr_mismatch) { 1597 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__); 1598 attrs.layer_etype = LAYER_MPA | DDP_LLP; 1599 attrs.ecode = MPA_NOMATCH_RTR; 1600 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1601 attrs.send_term = 1; 1602 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1603 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1604 err = -ENOMEM; 1605 disconnect = 1; 1606 goto out; 1607 } 1608 1609 /* 1610 * Generate TERM if initiator IRD is not sufficient for responder 1611 * provided ORD. Currently, we do the same behaviour even when 1612 * responder provided IRD is also not sufficient as regards to 1613 * initiator ORD. 1614 */ 1615 if (insuff_ird) { 1616 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n", 1617 __func__); 1618 attrs.layer_etype = LAYER_MPA | DDP_LLP; 1619 attrs.ecode = MPA_INSUFF_IRD; 1620 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1621 attrs.send_term = 1; 1622 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1623 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1624 err = -ENOMEM; 1625 disconnect = 1; 1626 goto out; 1627 } 1628 goto out; 1629 err_stop_timer: 1630 stop_ep_timer(ep); 1631 err: 1632 disconnect = 2; 1633 out: 1634 connect_reply_upcall(ep, err); 1635 return disconnect; 1636 } 1637 1638 /* 1639 * process_mpa_request - process streaming mode MPA request 1640 * 1641 * Returns: 1642 * 1643 * 0 upon success indicating a connect request was delivered to the ULP 1644 * or the mpa request is incomplete but valid so far. 1645 * 1646 * 1 if a failure requires the caller to close the connection. 1647 * 1648 * 2 if a failure requires the caller to abort the connection. 1649 */ 1650 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb) 1651 { 1652 struct mpa_message *mpa; 1653 struct mpa_v2_conn_params *mpa_v2_params; 1654 u16 plen; 1655 1656 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1657 1658 /* 1659 * If we get more than the supported amount of private data 1660 * then we must fail this connection. 1661 */ 1662 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) 1663 goto err_stop_timer; 1664 1665 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__); 1666 1667 /* 1668 * Copy the new data into our accumulation buffer. 1669 */ 1670 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]), 1671 skb->len); 1672 ep->mpa_pkt_len += skb->len; 1673 1674 /* 1675 * If we don't even have the mpa message, then bail. 1676 * We'll continue process when more data arrives. 1677 */ 1678 if (ep->mpa_pkt_len < sizeof(*mpa)) 1679 return 0; 1680 1681 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__); 1682 mpa = (struct mpa_message *) ep->mpa_pkt; 1683 1684 /* 1685 * Validate MPA Header. 1686 */ 1687 if (mpa->revision > mpa_rev) { 1688 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d," 1689 " Received = %d\n", __func__, mpa_rev, mpa->revision); 1690 goto err_stop_timer; 1691 } 1692 1693 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) 1694 goto err_stop_timer; 1695 1696 plen = ntohs(mpa->private_data_size); 1697 1698 /* 1699 * Fail if there's too much private data. 1700 */ 1701 if (plen > MPA_MAX_PRIVATE_DATA) 1702 goto err_stop_timer; 1703 1704 /* 1705 * If plen does not account for pkt size 1706 */ 1707 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) 1708 goto err_stop_timer; 1709 ep->plen = (u8) plen; 1710 1711 /* 1712 * If we don't have all the pdata yet, then bail. 1713 */ 1714 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 1715 return 0; 1716 1717 /* 1718 * If we get here we have accumulated the entire mpa 1719 * start reply message including private data. 1720 */ 1721 ep->mpa_attr.initiator = 0; 1722 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1723 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1724 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1725 ep->mpa_attr.version = mpa->revision; 1726 if (mpa->revision == 1) 1727 ep->tried_with_mpa_v1 = 1; 1728 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1729 1730 if (mpa->revision == 2) { 1731 ep->mpa_attr.enhanced_rdma_conn = 1732 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1733 if (ep->mpa_attr.enhanced_rdma_conn) { 1734 mpa_v2_params = (struct mpa_v2_conn_params *) 1735 (ep->mpa_pkt + sizeof(*mpa)); 1736 ep->ird = ntohs(mpa_v2_params->ird) & 1737 MPA_V2_IRD_ORD_MASK; 1738 ep->ird = min_t(u32, ep->ird, 1739 cur_max_read_depth(ep->com.dev)); 1740 ep->ord = ntohs(mpa_v2_params->ord) & 1741 MPA_V2_IRD_ORD_MASK; 1742 ep->ord = min_t(u32, ep->ord, 1743 cur_max_read_depth(ep->com.dev)); 1744 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird, 1745 ep->ord); 1746 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) 1747 if (peer2peer) { 1748 if (ntohs(mpa_v2_params->ord) & 1749 MPA_V2_RDMA_WRITE_RTR) 1750 ep->mpa_attr.p2p_type = 1751 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 1752 else if (ntohs(mpa_v2_params->ord) & 1753 MPA_V2_RDMA_READ_RTR) 1754 ep->mpa_attr.p2p_type = 1755 FW_RI_INIT_P2PTYPE_READ_REQ; 1756 } 1757 } 1758 } else if (mpa->revision == 1) 1759 if (peer2peer) 1760 ep->mpa_attr.p2p_type = p2p_type; 1761 1762 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, " 1763 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__, 1764 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 1765 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 1766 ep->mpa_attr.p2p_type); 1767 1768 __state_set(&ep->com, MPA_REQ_RCVD); 1769 1770 /* drive upcall */ 1771 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING); 1772 if (ep->parent_ep->com.state != DEAD) { 1773 if (connect_request_upcall(ep)) 1774 goto err_unlock_parent; 1775 } else { 1776 goto err_unlock_parent; 1777 } 1778 mutex_unlock(&ep->parent_ep->com.mutex); 1779 return 0; 1780 1781 err_unlock_parent: 1782 mutex_unlock(&ep->parent_ep->com.mutex); 1783 goto err_out; 1784 err_stop_timer: 1785 (void)stop_ep_timer(ep); 1786 err_out: 1787 return 2; 1788 } 1789 1790 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb) 1791 { 1792 struct c4iw_ep *ep; 1793 struct cpl_rx_data *hdr = cplhdr(skb); 1794 unsigned int dlen = ntohs(hdr->len); 1795 unsigned int tid = GET_TID(hdr); 1796 __u8 status = hdr->status; 1797 int disconnect = 0; 1798 1799 ep = get_ep_from_tid(dev, tid); 1800 if (!ep) 1801 return 0; 1802 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen); 1803 skb_pull(skb, sizeof(*hdr)); 1804 skb_trim(skb, dlen); 1805 mutex_lock(&ep->com.mutex); 1806 1807 switch (ep->com.state) { 1808 case MPA_REQ_SENT: 1809 update_rx_credits(ep, dlen); 1810 ep->rcv_seq += dlen; 1811 disconnect = process_mpa_reply(ep, skb); 1812 break; 1813 case MPA_REQ_WAIT: 1814 update_rx_credits(ep, dlen); 1815 ep->rcv_seq += dlen; 1816 disconnect = process_mpa_request(ep, skb); 1817 break; 1818 case FPDU_MODE: { 1819 struct c4iw_qp_attributes attrs; 1820 1821 update_rx_credits(ep, dlen); 1822 BUG_ON(!ep->com.qp); 1823 if (status) 1824 pr_err("%s Unexpected streaming data." \ 1825 " qpid %u ep %p state %d tid %u status %d\n", 1826 __func__, ep->com.qp->wq.sq.qid, ep, 1827 ep->com.state, ep->hwtid, status); 1828 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1829 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1830 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1831 disconnect = 1; 1832 break; 1833 } 1834 default: 1835 break; 1836 } 1837 mutex_unlock(&ep->com.mutex); 1838 if (disconnect) 1839 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL); 1840 c4iw_put_ep(&ep->com); 1841 return 0; 1842 } 1843 1844 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 1845 { 1846 struct c4iw_ep *ep; 1847 struct cpl_abort_rpl_rss *rpl = cplhdr(skb); 1848 int release = 0; 1849 unsigned int tid = GET_TID(rpl); 1850 1851 ep = get_ep_from_tid(dev, tid); 1852 if (!ep) { 1853 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n"); 1854 return 0; 1855 } 1856 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1857 mutex_lock(&ep->com.mutex); 1858 switch (ep->com.state) { 1859 case ABORTING: 1860 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 1861 __state_set(&ep->com, DEAD); 1862 release = 1; 1863 break; 1864 default: 1865 printk(KERN_ERR "%s ep %p state %d\n", 1866 __func__, ep, ep->com.state); 1867 break; 1868 } 1869 mutex_unlock(&ep->com.mutex); 1870 1871 if (release) 1872 release_ep_resources(ep); 1873 c4iw_put_ep(&ep->com); 1874 return 0; 1875 } 1876 1877 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid) 1878 { 1879 struct sk_buff *skb; 1880 struct fw_ofld_connection_wr *req; 1881 unsigned int mtu_idx; 1882 u32 wscale; 1883 struct sockaddr_in *sin; 1884 int win; 1885 1886 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL); 1887 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req)); 1888 memset(req, 0, sizeof(*req)); 1889 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR)); 1890 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16))); 1891 req->le.filter = cpu_to_be32(cxgb4_select_ntuple( 1892 ep->com.dev->rdev.lldi.ports[0], 1893 ep->l2t)); 1894 sin = (struct sockaddr_in *)&ep->com.local_addr; 1895 req->le.lport = sin->sin_port; 1896 req->le.u.ipv4.lip = sin->sin_addr.s_addr; 1897 sin = (struct sockaddr_in *)&ep->com.remote_addr; 1898 req->le.pport = sin->sin_port; 1899 req->le.u.ipv4.pip = sin->sin_addr.s_addr; 1900 req->tcb.t_state_to_astid = 1901 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) | 1902 FW_OFLD_CONNECTION_WR_ASTID_V(atid)); 1903 req->tcb.cplrxdataack_cplpassacceptrpl = 1904 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F); 1905 req->tcb.tx_max = (__force __be32) jiffies; 1906 req->tcb.rcv_adv = htons(1); 1907 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 1908 enable_tcp_timestamps, 1909 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1); 1910 wscale = cxgb_compute_wscale(rcv_win); 1911 1912 /* 1913 * Specify the largest window that will fit in opt0. The 1914 * remainder will be specified in the rx_data_ack. 1915 */ 1916 win = ep->rcv_win >> 10; 1917 if (win > RCV_BUFSIZ_M) 1918 win = RCV_BUFSIZ_M; 1919 1920 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F | 1921 (nocong ? NO_CONG_F : 0) | 1922 KEEP_ALIVE_F | 1923 DELACK_F | 1924 WND_SCALE_V(wscale) | 1925 MSS_IDX_V(mtu_idx) | 1926 L2T_IDX_V(ep->l2t->idx) | 1927 TX_CHAN_V(ep->tx_chan) | 1928 SMAC_SEL_V(ep->smac_idx) | 1929 DSCP_V(ep->tos >> 2) | 1930 ULP_MODE_V(ULP_MODE_TCPDDP) | 1931 RCV_BUFSIZ_V(win)); 1932 req->tcb.opt2 = (__force __be32) (PACE_V(1) | 1933 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) | 1934 RX_CHANNEL_V(0) | 1935 CCTRL_ECN_V(enable_ecn) | 1936 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid)); 1937 if (enable_tcp_timestamps) 1938 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F; 1939 if (enable_tcp_sack) 1940 req->tcb.opt2 |= (__force __be32)SACK_EN_F; 1941 if (wscale && enable_tcp_window_scaling) 1942 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F; 1943 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0); 1944 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2); 1945 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx); 1946 set_bit(ACT_OFLD_CONN, &ep->com.history); 1947 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1948 } 1949 1950 /* 1951 * Some of the error codes above implicitly indicate that there is no TID 1952 * allocated with the result of an ACT_OPEN. We use this predicate to make 1953 * that explicit. 1954 */ 1955 static inline int act_open_has_tid(int status) 1956 { 1957 return (status != CPL_ERR_TCAM_PARITY && 1958 status != CPL_ERR_TCAM_MISS && 1959 status != CPL_ERR_TCAM_FULL && 1960 status != CPL_ERR_CONN_EXIST_SYNRECV && 1961 status != CPL_ERR_CONN_EXIST); 1962 } 1963 1964 static char *neg_adv_str(unsigned int status) 1965 { 1966 switch (status) { 1967 case CPL_ERR_RTX_NEG_ADVICE: 1968 return "Retransmit timeout"; 1969 case CPL_ERR_PERSIST_NEG_ADVICE: 1970 return "Persist timeout"; 1971 case CPL_ERR_KEEPALV_NEG_ADVICE: 1972 return "Keepalive timeout"; 1973 default: 1974 return "Unknown"; 1975 } 1976 } 1977 1978 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi) 1979 { 1980 ep->snd_win = snd_win; 1981 ep->rcv_win = rcv_win; 1982 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win); 1983 } 1984 1985 #define ACT_OPEN_RETRY_COUNT 2 1986 1987 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip, 1988 struct dst_entry *dst, struct c4iw_dev *cdev, 1989 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos) 1990 { 1991 struct neighbour *n; 1992 int err, step; 1993 struct net_device *pdev; 1994 1995 n = dst_neigh_lookup(dst, peer_ip); 1996 if (!n) 1997 return -ENODEV; 1998 1999 rcu_read_lock(); 2000 err = -ENOMEM; 2001 if (n->dev->flags & IFF_LOOPBACK) { 2002 if (iptype == 4) 2003 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip); 2004 else if (IS_ENABLED(CONFIG_IPV6)) 2005 for_each_netdev(&init_net, pdev) { 2006 if (ipv6_chk_addr(&init_net, 2007 (struct in6_addr *)peer_ip, 2008 pdev, 1)) 2009 break; 2010 } 2011 else 2012 pdev = NULL; 2013 2014 if (!pdev) { 2015 err = -ENODEV; 2016 goto out; 2017 } 2018 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t, 2019 n, pdev, rt_tos2priority(tos)); 2020 if (!ep->l2t) { 2021 dev_put(pdev); 2022 goto out; 2023 } 2024 ep->mtu = pdev->mtu; 2025 ep->tx_chan = cxgb4_port_chan(pdev); 2026 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type, 2027 cxgb4_port_viid(pdev)); 2028 step = cdev->rdev.lldi.ntxq / 2029 cdev->rdev.lldi.nchan; 2030 ep->txq_idx = cxgb4_port_idx(pdev) * step; 2031 step = cdev->rdev.lldi.nrxq / 2032 cdev->rdev.lldi.nchan; 2033 ep->ctrlq_idx = cxgb4_port_idx(pdev); 2034 ep->rss_qid = cdev->rdev.lldi.rxq_ids[ 2035 cxgb4_port_idx(pdev) * step]; 2036 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev)); 2037 dev_put(pdev); 2038 } else { 2039 pdev = get_real_dev(n->dev); 2040 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t, 2041 n, pdev, 0); 2042 if (!ep->l2t) 2043 goto out; 2044 ep->mtu = dst_mtu(dst); 2045 ep->tx_chan = cxgb4_port_chan(pdev); 2046 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type, 2047 cxgb4_port_viid(pdev)); 2048 step = cdev->rdev.lldi.ntxq / 2049 cdev->rdev.lldi.nchan; 2050 ep->txq_idx = cxgb4_port_idx(pdev) * step; 2051 ep->ctrlq_idx = cxgb4_port_idx(pdev); 2052 step = cdev->rdev.lldi.nrxq / 2053 cdev->rdev.lldi.nchan; 2054 ep->rss_qid = cdev->rdev.lldi.rxq_ids[ 2055 cxgb4_port_idx(pdev) * step]; 2056 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev)); 2057 2058 if (clear_mpa_v1) { 2059 ep->retry_with_mpa_v1 = 0; 2060 ep->tried_with_mpa_v1 = 0; 2061 } 2062 } 2063 err = 0; 2064 out: 2065 rcu_read_unlock(); 2066 2067 neigh_release(n); 2068 2069 return err; 2070 } 2071 2072 static int c4iw_reconnect(struct c4iw_ep *ep) 2073 { 2074 int err = 0; 2075 int size = 0; 2076 struct sockaddr_in *laddr = (struct sockaddr_in *) 2077 &ep->com.cm_id->m_local_addr; 2078 struct sockaddr_in *raddr = (struct sockaddr_in *) 2079 &ep->com.cm_id->m_remote_addr; 2080 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *) 2081 &ep->com.cm_id->m_local_addr; 2082 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *) 2083 &ep->com.cm_id->m_remote_addr; 2084 int iptype; 2085 __u8 *ra; 2086 2087 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id); 2088 init_timer(&ep->timer); 2089 c4iw_init_wr_wait(&ep->com.wr_wait); 2090 2091 /* When MPA revision is different on nodes, the node with MPA_rev=2 2092 * tries to reconnect with MPA_rev 1 for the same EP through 2093 * c4iw_reconnect(), where the same EP is assigned with new tid for 2094 * further connection establishment. As we are using the same EP pointer 2095 * for reconnect, few skbs are used during the previous c4iw_connect(), 2096 * which leaves the EP with inadequate skbs for further 2097 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty 2098 * skb_list() during peer_abort(). Allocate skbs which is already used. 2099 */ 2100 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list)); 2101 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) { 2102 err = -ENOMEM; 2103 goto fail1; 2104 } 2105 2106 /* 2107 * Allocate an active TID to initiate a TCP connection. 2108 */ 2109 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep); 2110 if (ep->atid == -1) { 2111 pr_err("%s - cannot alloc atid.\n", __func__); 2112 err = -ENOMEM; 2113 goto fail2; 2114 } 2115 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid); 2116 2117 /* find a route */ 2118 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) { 2119 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev, 2120 laddr->sin_addr.s_addr, 2121 raddr->sin_addr.s_addr, 2122 laddr->sin_port, 2123 raddr->sin_port, ep->com.cm_id->tos); 2124 iptype = 4; 2125 ra = (__u8 *)&raddr->sin_addr; 2126 } else { 2127 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi, 2128 get_real_dev, 2129 laddr6->sin6_addr.s6_addr, 2130 raddr6->sin6_addr.s6_addr, 2131 laddr6->sin6_port, 2132 raddr6->sin6_port, 0, 2133 raddr6->sin6_scope_id); 2134 iptype = 6; 2135 ra = (__u8 *)&raddr6->sin6_addr; 2136 } 2137 if (!ep->dst) { 2138 pr_err("%s - cannot find route.\n", __func__); 2139 err = -EHOSTUNREACH; 2140 goto fail3; 2141 } 2142 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false, 2143 ep->com.dev->rdev.lldi.adapter_type, 2144 ep->com.cm_id->tos); 2145 if (err) { 2146 pr_err("%s - cannot alloc l2e.\n", __func__); 2147 goto fail4; 2148 } 2149 2150 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n", 2151 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid, 2152 ep->l2t->idx); 2153 2154 state_set(&ep->com, CONNECTING); 2155 ep->tos = ep->com.cm_id->tos; 2156 2157 /* send connect request to rnic */ 2158 err = send_connect(ep); 2159 if (!err) 2160 goto out; 2161 2162 cxgb4_l2t_release(ep->l2t); 2163 fail4: 2164 dst_release(ep->dst); 2165 fail3: 2166 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 2167 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 2168 fail2: 2169 /* 2170 * remember to send notification to upper layer. 2171 * We are in here so the upper layer is not aware that this is 2172 * re-connect attempt and so, upper layer is still waiting for 2173 * response of 1st connect request. 2174 */ 2175 connect_reply_upcall(ep, -ECONNRESET); 2176 fail1: 2177 c4iw_put_ep(&ep->com); 2178 out: 2179 return err; 2180 } 2181 2182 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2183 { 2184 struct c4iw_ep *ep; 2185 struct cpl_act_open_rpl *rpl = cplhdr(skb); 2186 unsigned int atid = TID_TID_G(AOPEN_ATID_G( 2187 ntohl(rpl->atid_status))); 2188 struct tid_info *t = dev->rdev.lldi.tids; 2189 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status)); 2190 struct sockaddr_in *la; 2191 struct sockaddr_in *ra; 2192 struct sockaddr_in6 *la6; 2193 struct sockaddr_in6 *ra6; 2194 int ret = 0; 2195 2196 ep = lookup_atid(t, atid); 2197 la = (struct sockaddr_in *)&ep->com.local_addr; 2198 ra = (struct sockaddr_in *)&ep->com.remote_addr; 2199 la6 = (struct sockaddr_in6 *)&ep->com.local_addr; 2200 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr; 2201 2202 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid, 2203 status, status2errno(status)); 2204 2205 if (cxgb_is_neg_adv(status)) { 2206 PDBG("%s Connection problems for atid %u status %u (%s)\n", 2207 __func__, atid, status, neg_adv_str(status)); 2208 ep->stats.connect_neg_adv++; 2209 mutex_lock(&dev->rdev.stats.lock); 2210 dev->rdev.stats.neg_adv++; 2211 mutex_unlock(&dev->rdev.stats.lock); 2212 return 0; 2213 } 2214 2215 set_bit(ACT_OPEN_RPL, &ep->com.history); 2216 2217 /* 2218 * Log interesting failures. 2219 */ 2220 switch (status) { 2221 case CPL_ERR_CONN_RESET: 2222 case CPL_ERR_CONN_TIMEDOUT: 2223 break; 2224 case CPL_ERR_TCAM_FULL: 2225 mutex_lock(&dev->rdev.stats.lock); 2226 dev->rdev.stats.tcam_full++; 2227 mutex_unlock(&dev->rdev.stats.lock); 2228 if (ep->com.local_addr.ss_family == AF_INET && 2229 dev->rdev.lldi.enable_fw_ofld_conn) { 2230 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G( 2231 ntohl(rpl->atid_status)))); 2232 if (ret) 2233 goto fail; 2234 return 0; 2235 } 2236 break; 2237 case CPL_ERR_CONN_EXIST: 2238 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 2239 set_bit(ACT_RETRY_INUSE, &ep->com.history); 2240 if (ep->com.remote_addr.ss_family == AF_INET6) { 2241 struct sockaddr_in6 *sin6 = 2242 (struct sockaddr_in6 *) 2243 &ep->com.local_addr; 2244 cxgb4_clip_release( 2245 ep->com.dev->rdev.lldi.ports[0], 2246 (const u32 *) 2247 &sin6->sin6_addr.s6_addr, 1); 2248 } 2249 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, 2250 atid); 2251 cxgb4_free_atid(t, atid); 2252 dst_release(ep->dst); 2253 cxgb4_l2t_release(ep->l2t); 2254 c4iw_reconnect(ep); 2255 return 0; 2256 } 2257 break; 2258 default: 2259 if (ep->com.local_addr.ss_family == AF_INET) { 2260 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n", 2261 atid, status, status2errno(status), 2262 &la->sin_addr.s_addr, ntohs(la->sin_port), 2263 &ra->sin_addr.s_addr, ntohs(ra->sin_port)); 2264 } else { 2265 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n", 2266 atid, status, status2errno(status), 2267 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port), 2268 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port)); 2269 } 2270 break; 2271 } 2272 2273 fail: 2274 connect_reply_upcall(ep, status2errno(status)); 2275 state_set(&ep->com, DEAD); 2276 2277 if (ep->com.remote_addr.ss_family == AF_INET6) { 2278 struct sockaddr_in6 *sin6 = 2279 (struct sockaddr_in6 *)&ep->com.local_addr; 2280 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 2281 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 2282 } 2283 if (status && act_open_has_tid(status)) 2284 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl)); 2285 2286 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid); 2287 cxgb4_free_atid(t, atid); 2288 dst_release(ep->dst); 2289 cxgb4_l2t_release(ep->l2t); 2290 c4iw_put_ep(&ep->com); 2291 2292 return 0; 2293 } 2294 2295 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2296 { 2297 struct cpl_pass_open_rpl *rpl = cplhdr(skb); 2298 unsigned int stid = GET_TID(rpl); 2299 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid); 2300 2301 if (!ep) { 2302 PDBG("%s stid %d lookup failure!\n", __func__, stid); 2303 goto out; 2304 } 2305 PDBG("%s ep %p status %d error %d\n", __func__, ep, 2306 rpl->status, status2errno(rpl->status)); 2307 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status)); 2308 c4iw_put_ep(&ep->com); 2309 out: 2310 return 0; 2311 } 2312 2313 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2314 { 2315 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb); 2316 unsigned int stid = GET_TID(rpl); 2317 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid); 2318 2319 PDBG("%s ep %p\n", __func__, ep); 2320 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status)); 2321 c4iw_put_ep(&ep->com); 2322 return 0; 2323 } 2324 2325 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb, 2326 struct cpl_pass_accept_req *req) 2327 { 2328 struct cpl_pass_accept_rpl *rpl; 2329 unsigned int mtu_idx; 2330 u64 opt0; 2331 u32 opt2; 2332 u32 wscale; 2333 struct cpl_t5_pass_accept_rpl *rpl5 = NULL; 2334 int win; 2335 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type; 2336 2337 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2338 BUG_ON(skb_cloned(skb)); 2339 2340 skb_get(skb); 2341 rpl = cplhdr(skb); 2342 if (!is_t4(adapter_type)) { 2343 skb_trim(skb, roundup(sizeof(*rpl5), 16)); 2344 rpl5 = (void *)rpl; 2345 INIT_TP_WR(rpl5, ep->hwtid); 2346 } else { 2347 skb_trim(skb, sizeof(*rpl)); 2348 INIT_TP_WR(rpl, ep->hwtid); 2349 } 2350 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, 2351 ep->hwtid)); 2352 2353 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 2354 enable_tcp_timestamps && req->tcpopt.tstamp, 2355 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1); 2356 wscale = cxgb_compute_wscale(rcv_win); 2357 2358 /* 2359 * Specify the largest window that will fit in opt0. The 2360 * remainder will be specified in the rx_data_ack. 2361 */ 2362 win = ep->rcv_win >> 10; 2363 if (win > RCV_BUFSIZ_M) 2364 win = RCV_BUFSIZ_M; 2365 opt0 = (nocong ? NO_CONG_F : 0) | 2366 KEEP_ALIVE_F | 2367 DELACK_F | 2368 WND_SCALE_V(wscale) | 2369 MSS_IDX_V(mtu_idx) | 2370 L2T_IDX_V(ep->l2t->idx) | 2371 TX_CHAN_V(ep->tx_chan) | 2372 SMAC_SEL_V(ep->smac_idx) | 2373 DSCP_V(ep->tos >> 2) | 2374 ULP_MODE_V(ULP_MODE_TCPDDP) | 2375 RCV_BUFSIZ_V(win); 2376 opt2 = RX_CHANNEL_V(0) | 2377 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid); 2378 2379 if (enable_tcp_timestamps && req->tcpopt.tstamp) 2380 opt2 |= TSTAMPS_EN_F; 2381 if (enable_tcp_sack && req->tcpopt.sack) 2382 opt2 |= SACK_EN_F; 2383 if (wscale && enable_tcp_window_scaling) 2384 opt2 |= WND_SCALE_EN_F; 2385 if (enable_ecn) { 2386 const struct tcphdr *tcph; 2387 u32 hlen = ntohl(req->hdr_len); 2388 2389 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5) 2390 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) + 2391 IP_HDR_LEN_G(hlen); 2392 else 2393 tcph = (const void *)(req + 1) + 2394 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen); 2395 if (tcph->ece && tcph->cwr) 2396 opt2 |= CCTRL_ECN_V(1); 2397 } 2398 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) { 2399 u32 isn = (prandom_u32() & ~7UL) - 1; 2400 opt2 |= T5_OPT_2_VALID_F; 2401 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE); 2402 opt2 |= T5_ISS_F; 2403 rpl5 = (void *)rpl; 2404 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16)); 2405 if (peer2peer) 2406 isn += 4; 2407 rpl5->iss = cpu_to_be32(isn); 2408 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss)); 2409 } 2410 2411 rpl->opt0 = cpu_to_be64(opt0); 2412 rpl->opt2 = cpu_to_be32(opt2); 2413 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx); 2414 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure); 2415 2416 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 2417 } 2418 2419 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb) 2420 { 2421 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid); 2422 BUG_ON(skb_cloned(skb)); 2423 skb_trim(skb, sizeof(struct cpl_tid_release)); 2424 release_tid(&dev->rdev, hwtid, skb); 2425 return; 2426 } 2427 2428 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb) 2429 { 2430 struct c4iw_ep *child_ep = NULL, *parent_ep; 2431 struct cpl_pass_accept_req *req = cplhdr(skb); 2432 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid)); 2433 struct tid_info *t = dev->rdev.lldi.tids; 2434 unsigned int hwtid = GET_TID(req); 2435 struct dst_entry *dst; 2436 __u8 local_ip[16], peer_ip[16]; 2437 __be16 local_port, peer_port; 2438 struct sockaddr_in6 *sin6; 2439 int err; 2440 u16 peer_mss = ntohs(req->tcpopt.mss); 2441 int iptype; 2442 unsigned short hdrs; 2443 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid)); 2444 2445 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid); 2446 if (!parent_ep) { 2447 PDBG("%s connect request on invalid stid %d\n", __func__, stid); 2448 goto reject; 2449 } 2450 2451 if (state_read(&parent_ep->com) != LISTEN) { 2452 PDBG("%s - listening ep not in LISTEN\n", __func__); 2453 goto reject; 2454 } 2455 2456 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, 2457 &iptype, local_ip, peer_ip, &local_port, &peer_port); 2458 2459 /* Find output route */ 2460 if (iptype == 4) { 2461 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n" 2462 , __func__, parent_ep, hwtid, 2463 local_ip, peer_ip, ntohs(local_port), 2464 ntohs(peer_port), peer_mss); 2465 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev, 2466 *(__be32 *)local_ip, *(__be32 *)peer_ip, 2467 local_port, peer_port, tos); 2468 } else { 2469 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n" 2470 , __func__, parent_ep, hwtid, 2471 local_ip, peer_ip, ntohs(local_port), 2472 ntohs(peer_port), peer_mss); 2473 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev, 2474 local_ip, peer_ip, local_port, peer_port, 2475 PASS_OPEN_TOS_G(ntohl(req->tos_stid)), 2476 ((struct sockaddr_in6 *) 2477 &parent_ep->com.local_addr)->sin6_scope_id); 2478 } 2479 if (!dst) { 2480 printk(KERN_ERR MOD "%s - failed to find dst entry!\n", 2481 __func__); 2482 goto reject; 2483 } 2484 2485 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL); 2486 if (!child_ep) { 2487 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n", 2488 __func__); 2489 dst_release(dst); 2490 goto reject; 2491 } 2492 2493 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false, 2494 parent_ep->com.dev->rdev.lldi.adapter_type, tos); 2495 if (err) { 2496 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n", 2497 __func__); 2498 dst_release(dst); 2499 kfree(child_ep); 2500 goto reject; 2501 } 2502 2503 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) + 2504 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0); 2505 if (peer_mss && child_ep->mtu > (peer_mss + hdrs)) 2506 child_ep->mtu = peer_mss + hdrs; 2507 2508 skb_queue_head_init(&child_ep->com.ep_skb_list); 2509 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF)) 2510 goto fail; 2511 2512 state_set(&child_ep->com, CONNECTING); 2513 child_ep->com.dev = dev; 2514 child_ep->com.cm_id = NULL; 2515 2516 if (iptype == 4) { 2517 struct sockaddr_in *sin = (struct sockaddr_in *) 2518 &child_ep->com.local_addr; 2519 2520 sin->sin_family = PF_INET; 2521 sin->sin_port = local_port; 2522 sin->sin_addr.s_addr = *(__be32 *)local_ip; 2523 2524 sin = (struct sockaddr_in *)&child_ep->com.local_addr; 2525 sin->sin_family = PF_INET; 2526 sin->sin_port = ((struct sockaddr_in *) 2527 &parent_ep->com.local_addr)->sin_port; 2528 sin->sin_addr.s_addr = *(__be32 *)local_ip; 2529 2530 sin = (struct sockaddr_in *)&child_ep->com.remote_addr; 2531 sin->sin_family = PF_INET; 2532 sin->sin_port = peer_port; 2533 sin->sin_addr.s_addr = *(__be32 *)peer_ip; 2534 } else { 2535 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2536 sin6->sin6_family = PF_INET6; 2537 sin6->sin6_port = local_port; 2538 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16); 2539 2540 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2541 sin6->sin6_family = PF_INET6; 2542 sin6->sin6_port = ((struct sockaddr_in6 *) 2543 &parent_ep->com.local_addr)->sin6_port; 2544 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16); 2545 2546 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr; 2547 sin6->sin6_family = PF_INET6; 2548 sin6->sin6_port = peer_port; 2549 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16); 2550 } 2551 2552 c4iw_get_ep(&parent_ep->com); 2553 child_ep->parent_ep = parent_ep; 2554 child_ep->tos = tos; 2555 child_ep->dst = dst; 2556 child_ep->hwtid = hwtid; 2557 2558 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__, 2559 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid); 2560 2561 init_timer(&child_ep->timer); 2562 cxgb4_insert_tid(t, child_ep, hwtid); 2563 insert_ep_tid(child_ep); 2564 if (accept_cr(child_ep, skb, req)) { 2565 c4iw_put_ep(&parent_ep->com); 2566 release_ep_resources(child_ep); 2567 } else { 2568 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history); 2569 } 2570 if (iptype == 6) { 2571 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2572 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0], 2573 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 2574 } 2575 goto out; 2576 fail: 2577 c4iw_put_ep(&child_ep->com); 2578 reject: 2579 reject_cr(dev, hwtid, skb); 2580 if (parent_ep) 2581 c4iw_put_ep(&parent_ep->com); 2582 out: 2583 return 0; 2584 } 2585 2586 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb) 2587 { 2588 struct c4iw_ep *ep; 2589 struct cpl_pass_establish *req = cplhdr(skb); 2590 unsigned int tid = GET_TID(req); 2591 int ret; 2592 2593 ep = get_ep_from_tid(dev, tid); 2594 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2595 ep->snd_seq = be32_to_cpu(req->snd_isn); 2596 ep->rcv_seq = be32_to_cpu(req->rcv_isn); 2597 2598 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid, 2599 ntohs(req->tcp_opt)); 2600 2601 set_emss(ep, ntohs(req->tcp_opt)); 2602 2603 dst_confirm(ep->dst); 2604 mutex_lock(&ep->com.mutex); 2605 ep->com.state = MPA_REQ_WAIT; 2606 start_ep_timer(ep); 2607 set_bit(PASS_ESTAB, &ep->com.history); 2608 ret = send_flowc(ep); 2609 mutex_unlock(&ep->com.mutex); 2610 if (ret) 2611 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 2612 c4iw_put_ep(&ep->com); 2613 2614 return 0; 2615 } 2616 2617 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb) 2618 { 2619 struct cpl_peer_close *hdr = cplhdr(skb); 2620 struct c4iw_ep *ep; 2621 struct c4iw_qp_attributes attrs; 2622 int disconnect = 1; 2623 int release = 0; 2624 unsigned int tid = GET_TID(hdr); 2625 int ret; 2626 2627 ep = get_ep_from_tid(dev, tid); 2628 if (!ep) 2629 return 0; 2630 2631 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2632 dst_confirm(ep->dst); 2633 2634 set_bit(PEER_CLOSE, &ep->com.history); 2635 mutex_lock(&ep->com.mutex); 2636 switch (ep->com.state) { 2637 case MPA_REQ_WAIT: 2638 __state_set(&ep->com, CLOSING); 2639 break; 2640 case MPA_REQ_SENT: 2641 __state_set(&ep->com, CLOSING); 2642 connect_reply_upcall(ep, -ECONNRESET); 2643 break; 2644 case MPA_REQ_RCVD: 2645 2646 /* 2647 * We're gonna mark this puppy DEAD, but keep 2648 * the reference on it until the ULP accepts or 2649 * rejects the CR. Also wake up anyone waiting 2650 * in rdma connection migration (see c4iw_accept_cr()). 2651 */ 2652 __state_set(&ep->com, CLOSING); 2653 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid); 2654 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2655 break; 2656 case MPA_REP_SENT: 2657 __state_set(&ep->com, CLOSING); 2658 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid); 2659 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2660 break; 2661 case FPDU_MODE: 2662 start_ep_timer(ep); 2663 __state_set(&ep->com, CLOSING); 2664 attrs.next_state = C4IW_QP_STATE_CLOSING; 2665 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2666 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2667 if (ret != -ECONNRESET) { 2668 peer_close_upcall(ep); 2669 disconnect = 1; 2670 } 2671 break; 2672 case ABORTING: 2673 disconnect = 0; 2674 break; 2675 case CLOSING: 2676 __state_set(&ep->com, MORIBUND); 2677 disconnect = 0; 2678 break; 2679 case MORIBUND: 2680 (void)stop_ep_timer(ep); 2681 if (ep->com.cm_id && ep->com.qp) { 2682 attrs.next_state = C4IW_QP_STATE_IDLE; 2683 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2684 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2685 } 2686 close_complete_upcall(ep, 0); 2687 __state_set(&ep->com, DEAD); 2688 release = 1; 2689 disconnect = 0; 2690 break; 2691 case DEAD: 2692 disconnect = 0; 2693 break; 2694 default: 2695 BUG_ON(1); 2696 } 2697 mutex_unlock(&ep->com.mutex); 2698 if (disconnect) 2699 c4iw_ep_disconnect(ep, 0, GFP_KERNEL); 2700 if (release) 2701 release_ep_resources(ep); 2702 c4iw_put_ep(&ep->com); 2703 return 0; 2704 } 2705 2706 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb) 2707 { 2708 struct cpl_abort_req_rss *req = cplhdr(skb); 2709 struct c4iw_ep *ep; 2710 struct sk_buff *rpl_skb; 2711 struct c4iw_qp_attributes attrs; 2712 int ret; 2713 int release = 0; 2714 unsigned int tid = GET_TID(req); 2715 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16); 2716 2717 ep = get_ep_from_tid(dev, tid); 2718 if (!ep) 2719 return 0; 2720 2721 if (cxgb_is_neg_adv(req->status)) { 2722 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n", 2723 __func__, ep->hwtid, req->status, 2724 neg_adv_str(req->status)); 2725 ep->stats.abort_neg_adv++; 2726 mutex_lock(&dev->rdev.stats.lock); 2727 dev->rdev.stats.neg_adv++; 2728 mutex_unlock(&dev->rdev.stats.lock); 2729 goto deref_ep; 2730 } 2731 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid, 2732 ep->com.state); 2733 set_bit(PEER_ABORT, &ep->com.history); 2734 2735 /* 2736 * Wake up any threads in rdma_init() or rdma_fini(). 2737 * However, this is not needed if com state is just 2738 * MPA_REQ_SENT 2739 */ 2740 if (ep->com.state != MPA_REQ_SENT) 2741 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2742 2743 mutex_lock(&ep->com.mutex); 2744 switch (ep->com.state) { 2745 case CONNECTING: 2746 c4iw_put_ep(&ep->parent_ep->com); 2747 break; 2748 case MPA_REQ_WAIT: 2749 (void)stop_ep_timer(ep); 2750 break; 2751 case MPA_REQ_SENT: 2752 (void)stop_ep_timer(ep); 2753 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1)) 2754 connect_reply_upcall(ep, -ECONNRESET); 2755 else { 2756 /* 2757 * we just don't send notification upwards because we 2758 * want to retry with mpa_v1 without upper layers even 2759 * knowing it. 2760 * 2761 * do some housekeeping so as to re-initiate the 2762 * connection 2763 */ 2764 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__, 2765 mpa_rev); 2766 ep->retry_with_mpa_v1 = 1; 2767 } 2768 break; 2769 case MPA_REP_SENT: 2770 break; 2771 case MPA_REQ_RCVD: 2772 break; 2773 case MORIBUND: 2774 case CLOSING: 2775 stop_ep_timer(ep); 2776 /*FALLTHROUGH*/ 2777 case FPDU_MODE: 2778 if (ep->com.cm_id && ep->com.qp) { 2779 attrs.next_state = C4IW_QP_STATE_ERROR; 2780 ret = c4iw_modify_qp(ep->com.qp->rhp, 2781 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 2782 &attrs, 1); 2783 if (ret) 2784 printk(KERN_ERR MOD 2785 "%s - qp <- error failed!\n", 2786 __func__); 2787 } 2788 peer_abort_upcall(ep); 2789 break; 2790 case ABORTING: 2791 break; 2792 case DEAD: 2793 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__); 2794 mutex_unlock(&ep->com.mutex); 2795 goto deref_ep; 2796 default: 2797 BUG_ON(1); 2798 break; 2799 } 2800 dst_confirm(ep->dst); 2801 if (ep->com.state != ABORTING) { 2802 __state_set(&ep->com, DEAD); 2803 /* we don't release if we want to retry with mpa_v1 */ 2804 if (!ep->retry_with_mpa_v1) 2805 release = 1; 2806 } 2807 mutex_unlock(&ep->com.mutex); 2808 2809 rpl_skb = skb_dequeue(&ep->com.ep_skb_list); 2810 if (WARN_ON(!rpl_skb)) { 2811 release = 1; 2812 goto out; 2813 } 2814 2815 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx); 2816 2817 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb); 2818 out: 2819 if (release) 2820 release_ep_resources(ep); 2821 else if (ep->retry_with_mpa_v1) { 2822 if (ep->com.remote_addr.ss_family == AF_INET6) { 2823 struct sockaddr_in6 *sin6 = 2824 (struct sockaddr_in6 *) 2825 &ep->com.local_addr; 2826 cxgb4_clip_release( 2827 ep->com.dev->rdev.lldi.ports[0], 2828 (const u32 *)&sin6->sin6_addr.s6_addr, 2829 1); 2830 } 2831 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid); 2832 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid); 2833 dst_release(ep->dst); 2834 cxgb4_l2t_release(ep->l2t); 2835 c4iw_reconnect(ep); 2836 } 2837 2838 deref_ep: 2839 c4iw_put_ep(&ep->com); 2840 /* Dereferencing ep, referenced in peer_abort_intr() */ 2841 c4iw_put_ep(&ep->com); 2842 return 0; 2843 } 2844 2845 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2846 { 2847 struct c4iw_ep *ep; 2848 struct c4iw_qp_attributes attrs; 2849 struct cpl_close_con_rpl *rpl = cplhdr(skb); 2850 int release = 0; 2851 unsigned int tid = GET_TID(rpl); 2852 2853 ep = get_ep_from_tid(dev, tid); 2854 if (!ep) 2855 return 0; 2856 2857 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2858 BUG_ON(!ep); 2859 2860 /* The cm_id may be null if we failed to connect */ 2861 mutex_lock(&ep->com.mutex); 2862 set_bit(CLOSE_CON_RPL, &ep->com.history); 2863 switch (ep->com.state) { 2864 case CLOSING: 2865 __state_set(&ep->com, MORIBUND); 2866 break; 2867 case MORIBUND: 2868 (void)stop_ep_timer(ep); 2869 if ((ep->com.cm_id) && (ep->com.qp)) { 2870 attrs.next_state = C4IW_QP_STATE_IDLE; 2871 c4iw_modify_qp(ep->com.qp->rhp, 2872 ep->com.qp, 2873 C4IW_QP_ATTR_NEXT_STATE, 2874 &attrs, 1); 2875 } 2876 close_complete_upcall(ep, 0); 2877 __state_set(&ep->com, DEAD); 2878 release = 1; 2879 break; 2880 case ABORTING: 2881 case DEAD: 2882 break; 2883 default: 2884 BUG_ON(1); 2885 break; 2886 } 2887 mutex_unlock(&ep->com.mutex); 2888 if (release) 2889 release_ep_resources(ep); 2890 c4iw_put_ep(&ep->com); 2891 return 0; 2892 } 2893 2894 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb) 2895 { 2896 struct cpl_rdma_terminate *rpl = cplhdr(skb); 2897 unsigned int tid = GET_TID(rpl); 2898 struct c4iw_ep *ep; 2899 struct c4iw_qp_attributes attrs; 2900 2901 ep = get_ep_from_tid(dev, tid); 2902 BUG_ON(!ep); 2903 2904 if (ep && ep->com.qp) { 2905 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid, 2906 ep->com.qp->wq.sq.qid); 2907 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2908 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2909 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2910 } else 2911 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid); 2912 c4iw_put_ep(&ep->com); 2913 2914 return 0; 2915 } 2916 2917 /* 2918 * Upcall from the adapter indicating data has been transmitted. 2919 * For us its just the single MPA request or reply. We can now free 2920 * the skb holding the mpa message. 2921 */ 2922 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb) 2923 { 2924 struct c4iw_ep *ep; 2925 struct cpl_fw4_ack *hdr = cplhdr(skb); 2926 u8 credits = hdr->credits; 2927 unsigned int tid = GET_TID(hdr); 2928 2929 2930 ep = get_ep_from_tid(dev, tid); 2931 if (!ep) 2932 return 0; 2933 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits); 2934 if (credits == 0) { 2935 PDBG("%s 0 credit ack ep %p tid %u state %u\n", 2936 __func__, ep, ep->hwtid, state_read(&ep->com)); 2937 goto out; 2938 } 2939 2940 dst_confirm(ep->dst); 2941 if (ep->mpa_skb) { 2942 PDBG("%s last streaming msg ack ep %p tid %u state %u " 2943 "initiator %u freeing skb\n", __func__, ep, ep->hwtid, 2944 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0); 2945 mutex_lock(&ep->com.mutex); 2946 kfree_skb(ep->mpa_skb); 2947 ep->mpa_skb = NULL; 2948 if (test_bit(STOP_MPA_TIMER, &ep->com.flags)) 2949 stop_ep_timer(ep); 2950 mutex_unlock(&ep->com.mutex); 2951 } 2952 out: 2953 c4iw_put_ep(&ep->com); 2954 return 0; 2955 } 2956 2957 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) 2958 { 2959 int abort; 2960 struct c4iw_ep *ep = to_ep(cm_id); 2961 2962 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2963 2964 mutex_lock(&ep->com.mutex); 2965 if (ep->com.state != MPA_REQ_RCVD) { 2966 mutex_unlock(&ep->com.mutex); 2967 c4iw_put_ep(&ep->com); 2968 return -ECONNRESET; 2969 } 2970 set_bit(ULP_REJECT, &ep->com.history); 2971 if (mpa_rev == 0) 2972 abort = 1; 2973 else 2974 abort = send_mpa_reject(ep, pdata, pdata_len); 2975 mutex_unlock(&ep->com.mutex); 2976 2977 stop_ep_timer(ep); 2978 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL); 2979 c4iw_put_ep(&ep->com); 2980 return 0; 2981 } 2982 2983 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2984 { 2985 int err; 2986 struct c4iw_qp_attributes attrs; 2987 enum c4iw_qp_attr_mask mask; 2988 struct c4iw_ep *ep = to_ep(cm_id); 2989 struct c4iw_dev *h = to_c4iw_dev(cm_id->device); 2990 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn); 2991 int abort = 0; 2992 2993 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2994 2995 mutex_lock(&ep->com.mutex); 2996 if (ep->com.state != MPA_REQ_RCVD) { 2997 err = -ECONNRESET; 2998 goto err_out; 2999 } 3000 3001 BUG_ON(!qp); 3002 3003 set_bit(ULP_ACCEPT, &ep->com.history); 3004 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) || 3005 (conn_param->ird > cur_max_read_depth(ep->com.dev))) { 3006 err = -EINVAL; 3007 goto err_abort; 3008 } 3009 3010 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 3011 if (conn_param->ord > ep->ird) { 3012 if (RELAXED_IRD_NEGOTIATION) { 3013 conn_param->ord = ep->ird; 3014 } else { 3015 ep->ird = conn_param->ird; 3016 ep->ord = conn_param->ord; 3017 send_mpa_reject(ep, conn_param->private_data, 3018 conn_param->private_data_len); 3019 err = -ENOMEM; 3020 goto err_abort; 3021 } 3022 } 3023 if (conn_param->ird < ep->ord) { 3024 if (RELAXED_IRD_NEGOTIATION && 3025 ep->ord <= h->rdev.lldi.max_ordird_qp) { 3026 conn_param->ird = ep->ord; 3027 } else { 3028 err = -ENOMEM; 3029 goto err_abort; 3030 } 3031 } 3032 } 3033 ep->ird = conn_param->ird; 3034 ep->ord = conn_param->ord; 3035 3036 if (ep->mpa_attr.version == 1) { 3037 if (peer2peer && ep->ird == 0) 3038 ep->ird = 1; 3039 } else { 3040 if (peer2peer && 3041 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) && 3042 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0) 3043 ep->ird = 1; 3044 } 3045 3046 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord); 3047 3048 ep->com.cm_id = cm_id; 3049 ref_cm_id(&ep->com); 3050 ep->com.qp = qp; 3051 ref_qp(ep); 3052 3053 /* bind QP to EP and move to RTS */ 3054 attrs.mpa_attr = ep->mpa_attr; 3055 attrs.max_ird = ep->ird; 3056 attrs.max_ord = ep->ord; 3057 attrs.llp_stream_handle = ep; 3058 attrs.next_state = C4IW_QP_STATE_RTS; 3059 3060 /* bind QP and TID with INIT_WR */ 3061 mask = C4IW_QP_ATTR_NEXT_STATE | 3062 C4IW_QP_ATTR_LLP_STREAM_HANDLE | 3063 C4IW_QP_ATTR_MPA_ATTR | 3064 C4IW_QP_ATTR_MAX_IRD | 3065 C4IW_QP_ATTR_MAX_ORD; 3066 3067 err = c4iw_modify_qp(ep->com.qp->rhp, 3068 ep->com.qp, mask, &attrs, 1); 3069 if (err) 3070 goto err_deref_cm_id; 3071 3072 set_bit(STOP_MPA_TIMER, &ep->com.flags); 3073 err = send_mpa_reply(ep, conn_param->private_data, 3074 conn_param->private_data_len); 3075 if (err) 3076 goto err_deref_cm_id; 3077 3078 __state_set(&ep->com, FPDU_MODE); 3079 established_upcall(ep); 3080 mutex_unlock(&ep->com.mutex); 3081 c4iw_put_ep(&ep->com); 3082 return 0; 3083 err_deref_cm_id: 3084 deref_cm_id(&ep->com); 3085 err_abort: 3086 abort = 1; 3087 err_out: 3088 mutex_unlock(&ep->com.mutex); 3089 if (abort) 3090 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 3091 c4iw_put_ep(&ep->com); 3092 return err; 3093 } 3094 3095 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id) 3096 { 3097 struct in_device *ind; 3098 int found = 0; 3099 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr; 3100 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr; 3101 3102 ind = in_dev_get(dev->rdev.lldi.ports[0]); 3103 if (!ind) 3104 return -EADDRNOTAVAIL; 3105 for_primary_ifa(ind) { 3106 laddr->sin_addr.s_addr = ifa->ifa_address; 3107 raddr->sin_addr.s_addr = ifa->ifa_address; 3108 found = 1; 3109 break; 3110 } 3111 endfor_ifa(ind); 3112 in_dev_put(ind); 3113 return found ? 0 : -EADDRNOTAVAIL; 3114 } 3115 3116 static int get_lladdr(struct net_device *dev, struct in6_addr *addr, 3117 unsigned char banned_flags) 3118 { 3119 struct inet6_dev *idev; 3120 int err = -EADDRNOTAVAIL; 3121 3122 rcu_read_lock(); 3123 idev = __in6_dev_get(dev); 3124 if (idev != NULL) { 3125 struct inet6_ifaddr *ifp; 3126 3127 read_lock_bh(&idev->lock); 3128 list_for_each_entry(ifp, &idev->addr_list, if_list) { 3129 if (ifp->scope == IFA_LINK && 3130 !(ifp->flags & banned_flags)) { 3131 memcpy(addr, &ifp->addr, 16); 3132 err = 0; 3133 break; 3134 } 3135 } 3136 read_unlock_bh(&idev->lock); 3137 } 3138 rcu_read_unlock(); 3139 return err; 3140 } 3141 3142 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id) 3143 { 3144 struct in6_addr uninitialized_var(addr); 3145 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr; 3146 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr; 3147 3148 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) { 3149 memcpy(la6->sin6_addr.s6_addr, &addr, 16); 3150 memcpy(ra6->sin6_addr.s6_addr, &addr, 16); 3151 return 0; 3152 } 3153 return -EADDRNOTAVAIL; 3154 } 3155 3156 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 3157 { 3158 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 3159 struct c4iw_ep *ep; 3160 int err = 0; 3161 struct sockaddr_in *laddr; 3162 struct sockaddr_in *raddr; 3163 struct sockaddr_in6 *laddr6; 3164 struct sockaddr_in6 *raddr6; 3165 __u8 *ra; 3166 int iptype; 3167 3168 if ((conn_param->ord > cur_max_read_depth(dev)) || 3169 (conn_param->ird > cur_max_read_depth(dev))) { 3170 err = -EINVAL; 3171 goto out; 3172 } 3173 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 3174 if (!ep) { 3175 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__); 3176 err = -ENOMEM; 3177 goto out; 3178 } 3179 3180 skb_queue_head_init(&ep->com.ep_skb_list); 3181 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) { 3182 err = -ENOMEM; 3183 goto fail1; 3184 } 3185 3186 init_timer(&ep->timer); 3187 ep->plen = conn_param->private_data_len; 3188 if (ep->plen) 3189 memcpy(ep->mpa_pkt + sizeof(struct mpa_message), 3190 conn_param->private_data, ep->plen); 3191 ep->ird = conn_param->ird; 3192 ep->ord = conn_param->ord; 3193 3194 if (peer2peer && ep->ord == 0) 3195 ep->ord = 1; 3196 3197 ep->com.cm_id = cm_id; 3198 ref_cm_id(&ep->com); 3199 ep->com.dev = dev; 3200 ep->com.qp = get_qhp(dev, conn_param->qpn); 3201 if (!ep->com.qp) { 3202 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn); 3203 err = -EINVAL; 3204 goto fail2; 3205 } 3206 ref_qp(ep); 3207 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn, 3208 ep->com.qp, cm_id); 3209 3210 /* 3211 * Allocate an active TID to initiate a TCP connection. 3212 */ 3213 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep); 3214 if (ep->atid == -1) { 3215 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__); 3216 err = -ENOMEM; 3217 goto fail2; 3218 } 3219 insert_handle(dev, &dev->atid_idr, ep, ep->atid); 3220 3221 memcpy(&ep->com.local_addr, &cm_id->m_local_addr, 3222 sizeof(ep->com.local_addr)); 3223 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr, 3224 sizeof(ep->com.remote_addr)); 3225 3226 laddr = (struct sockaddr_in *)&ep->com.local_addr; 3227 raddr = (struct sockaddr_in *)&ep->com.remote_addr; 3228 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr; 3229 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr; 3230 3231 if (cm_id->m_remote_addr.ss_family == AF_INET) { 3232 iptype = 4; 3233 ra = (__u8 *)&raddr->sin_addr; 3234 3235 /* 3236 * Handle loopback requests to INADDR_ANY. 3237 */ 3238 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) { 3239 err = pick_local_ipaddrs(dev, cm_id); 3240 if (err) 3241 goto fail2; 3242 } 3243 3244 /* find a route */ 3245 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n", 3246 __func__, &laddr->sin_addr, ntohs(laddr->sin_port), 3247 ra, ntohs(raddr->sin_port)); 3248 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev, 3249 laddr->sin_addr.s_addr, 3250 raddr->sin_addr.s_addr, 3251 laddr->sin_port, 3252 raddr->sin_port, cm_id->tos); 3253 } else { 3254 iptype = 6; 3255 ra = (__u8 *)&raddr6->sin6_addr; 3256 3257 /* 3258 * Handle loopback requests to INADDR_ANY. 3259 */ 3260 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) { 3261 err = pick_local_ip6addrs(dev, cm_id); 3262 if (err) 3263 goto fail2; 3264 } 3265 3266 /* find a route */ 3267 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n", 3268 __func__, laddr6->sin6_addr.s6_addr, 3269 ntohs(laddr6->sin6_port), 3270 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port)); 3271 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev, 3272 laddr6->sin6_addr.s6_addr, 3273 raddr6->sin6_addr.s6_addr, 3274 laddr6->sin6_port, 3275 raddr6->sin6_port, 0, 3276 raddr6->sin6_scope_id); 3277 } 3278 if (!ep->dst) { 3279 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__); 3280 err = -EHOSTUNREACH; 3281 goto fail3; 3282 } 3283 3284 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true, 3285 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos); 3286 if (err) { 3287 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__); 3288 goto fail4; 3289 } 3290 3291 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n", 3292 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid, 3293 ep->l2t->idx); 3294 3295 state_set(&ep->com, CONNECTING); 3296 ep->tos = cm_id->tos; 3297 3298 /* send connect request to rnic */ 3299 err = send_connect(ep); 3300 if (!err) 3301 goto out; 3302 3303 cxgb4_l2t_release(ep->l2t); 3304 fail4: 3305 dst_release(ep->dst); 3306 fail3: 3307 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 3308 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 3309 fail2: 3310 skb_queue_purge(&ep->com.ep_skb_list); 3311 deref_cm_id(&ep->com); 3312 fail1: 3313 c4iw_put_ep(&ep->com); 3314 out: 3315 return err; 3316 } 3317 3318 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) 3319 { 3320 int err; 3321 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) 3322 &ep->com.local_addr; 3323 3324 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) { 3325 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0], 3326 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3327 if (err) 3328 return err; 3329 } 3330 c4iw_init_wr_wait(&ep->com.wr_wait); 3331 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0], 3332 ep->stid, &sin6->sin6_addr, 3333 sin6->sin6_port, 3334 ep->com.dev->rdev.lldi.rxq_ids[0]); 3335 if (!err) 3336 err = c4iw_wait_for_reply(&ep->com.dev->rdev, 3337 &ep->com.wr_wait, 3338 0, 0, __func__); 3339 else if (err > 0) 3340 err = net_xmit_errno(err); 3341 if (err) { 3342 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3343 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3344 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n", 3345 err, ep->stid, 3346 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port)); 3347 } 3348 return err; 3349 } 3350 3351 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) 3352 { 3353 int err; 3354 struct sockaddr_in *sin = (struct sockaddr_in *) 3355 &ep->com.local_addr; 3356 3357 if (dev->rdev.lldi.enable_fw_ofld_conn) { 3358 do { 3359 err = cxgb4_create_server_filter( 3360 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3361 sin->sin_addr.s_addr, sin->sin_port, 0, 3362 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0); 3363 if (err == -EBUSY) { 3364 if (c4iw_fatal_error(&ep->com.dev->rdev)) { 3365 err = -EIO; 3366 break; 3367 } 3368 set_current_state(TASK_UNINTERRUPTIBLE); 3369 schedule_timeout(usecs_to_jiffies(100)); 3370 } 3371 } while (err == -EBUSY); 3372 } else { 3373 c4iw_init_wr_wait(&ep->com.wr_wait); 3374 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0], 3375 ep->stid, sin->sin_addr.s_addr, sin->sin_port, 3376 0, ep->com.dev->rdev.lldi.rxq_ids[0]); 3377 if (!err) 3378 err = c4iw_wait_for_reply(&ep->com.dev->rdev, 3379 &ep->com.wr_wait, 3380 0, 0, __func__); 3381 else if (err > 0) 3382 err = net_xmit_errno(err); 3383 } 3384 if (err) 3385 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n" 3386 , err, ep->stid, 3387 &sin->sin_addr, ntohs(sin->sin_port)); 3388 return err; 3389 } 3390 3391 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) 3392 { 3393 int err = 0; 3394 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 3395 struct c4iw_listen_ep *ep; 3396 3397 might_sleep(); 3398 3399 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 3400 if (!ep) { 3401 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__); 3402 err = -ENOMEM; 3403 goto fail1; 3404 } 3405 skb_queue_head_init(&ep->com.ep_skb_list); 3406 PDBG("%s ep %p\n", __func__, ep); 3407 ep->com.cm_id = cm_id; 3408 ref_cm_id(&ep->com); 3409 ep->com.dev = dev; 3410 ep->backlog = backlog; 3411 memcpy(&ep->com.local_addr, &cm_id->m_local_addr, 3412 sizeof(ep->com.local_addr)); 3413 3414 /* 3415 * Allocate a server TID. 3416 */ 3417 if (dev->rdev.lldi.enable_fw_ofld_conn && 3418 ep->com.local_addr.ss_family == AF_INET) 3419 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids, 3420 cm_id->m_local_addr.ss_family, ep); 3421 else 3422 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, 3423 cm_id->m_local_addr.ss_family, ep); 3424 3425 if (ep->stid == -1) { 3426 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__); 3427 err = -ENOMEM; 3428 goto fail2; 3429 } 3430 insert_handle(dev, &dev->stid_idr, ep, ep->stid); 3431 3432 memcpy(&ep->com.local_addr, &cm_id->m_local_addr, 3433 sizeof(ep->com.local_addr)); 3434 3435 state_set(&ep->com, LISTEN); 3436 if (ep->com.local_addr.ss_family == AF_INET) 3437 err = create_server4(dev, ep); 3438 else 3439 err = create_server6(dev, ep); 3440 if (!err) { 3441 cm_id->provider_data = ep; 3442 goto out; 3443 } 3444 3445 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, 3446 ep->com.local_addr.ss_family); 3447 fail2: 3448 deref_cm_id(&ep->com); 3449 c4iw_put_ep(&ep->com); 3450 fail1: 3451 out: 3452 return err; 3453 } 3454 3455 int c4iw_destroy_listen(struct iw_cm_id *cm_id) 3456 { 3457 int err; 3458 struct c4iw_listen_ep *ep = to_listen_ep(cm_id); 3459 3460 PDBG("%s ep %p\n", __func__, ep); 3461 3462 might_sleep(); 3463 state_set(&ep->com, DEAD); 3464 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn && 3465 ep->com.local_addr.ss_family == AF_INET) { 3466 err = cxgb4_remove_server_filter( 3467 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3468 ep->com.dev->rdev.lldi.rxq_ids[0], 0); 3469 } else { 3470 struct sockaddr_in6 *sin6; 3471 c4iw_init_wr_wait(&ep->com.wr_wait); 3472 err = cxgb4_remove_server( 3473 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3474 ep->com.dev->rdev.lldi.rxq_ids[0], 0); 3475 if (err) 3476 goto done; 3477 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 3478 0, 0, __func__); 3479 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr; 3480 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3481 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3482 } 3483 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid); 3484 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, 3485 ep->com.local_addr.ss_family); 3486 done: 3487 deref_cm_id(&ep->com); 3488 c4iw_put_ep(&ep->com); 3489 return err; 3490 } 3491 3492 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 3493 { 3494 int ret = 0; 3495 int close = 0; 3496 int fatal = 0; 3497 struct c4iw_rdev *rdev; 3498 3499 mutex_lock(&ep->com.mutex); 3500 3501 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep, 3502 states[ep->com.state], abrupt); 3503 3504 /* 3505 * Ref the ep here in case we have fatal errors causing the 3506 * ep to be released and freed. 3507 */ 3508 c4iw_get_ep(&ep->com); 3509 3510 rdev = &ep->com.dev->rdev; 3511 if (c4iw_fatal_error(rdev)) { 3512 fatal = 1; 3513 close_complete_upcall(ep, -EIO); 3514 ep->com.state = DEAD; 3515 } 3516 switch (ep->com.state) { 3517 case MPA_REQ_WAIT: 3518 case MPA_REQ_SENT: 3519 case MPA_REQ_RCVD: 3520 case MPA_REP_SENT: 3521 case FPDU_MODE: 3522 case CONNECTING: 3523 close = 1; 3524 if (abrupt) 3525 ep->com.state = ABORTING; 3526 else { 3527 ep->com.state = CLOSING; 3528 3529 /* 3530 * if we close before we see the fw4_ack() then we fix 3531 * up the timer state since we're reusing it. 3532 */ 3533 if (ep->mpa_skb && 3534 test_bit(STOP_MPA_TIMER, &ep->com.flags)) { 3535 clear_bit(STOP_MPA_TIMER, &ep->com.flags); 3536 stop_ep_timer(ep); 3537 } 3538 start_ep_timer(ep); 3539 } 3540 set_bit(CLOSE_SENT, &ep->com.flags); 3541 break; 3542 case CLOSING: 3543 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) { 3544 close = 1; 3545 if (abrupt) { 3546 (void)stop_ep_timer(ep); 3547 ep->com.state = ABORTING; 3548 } else 3549 ep->com.state = MORIBUND; 3550 } 3551 break; 3552 case MORIBUND: 3553 case ABORTING: 3554 case DEAD: 3555 PDBG("%s ignoring disconnect ep %p state %u\n", 3556 __func__, ep, ep->com.state); 3557 break; 3558 default: 3559 BUG(); 3560 break; 3561 } 3562 3563 if (close) { 3564 if (abrupt) { 3565 set_bit(EP_DISC_ABORT, &ep->com.history); 3566 close_complete_upcall(ep, -ECONNRESET); 3567 ret = send_abort(ep); 3568 } else { 3569 set_bit(EP_DISC_CLOSE, &ep->com.history); 3570 ret = send_halfclose(ep); 3571 } 3572 if (ret) { 3573 set_bit(EP_DISC_FAIL, &ep->com.history); 3574 if (!abrupt) { 3575 stop_ep_timer(ep); 3576 close_complete_upcall(ep, -EIO); 3577 } 3578 if (ep->com.qp) { 3579 struct c4iw_qp_attributes attrs; 3580 3581 attrs.next_state = C4IW_QP_STATE_ERROR; 3582 ret = c4iw_modify_qp(ep->com.qp->rhp, 3583 ep->com.qp, 3584 C4IW_QP_ATTR_NEXT_STATE, 3585 &attrs, 1); 3586 if (ret) 3587 pr_err(MOD 3588 "%s - qp <- error failed!\n", 3589 __func__); 3590 } 3591 fatal = 1; 3592 } 3593 } 3594 mutex_unlock(&ep->com.mutex); 3595 c4iw_put_ep(&ep->com); 3596 if (fatal) 3597 release_ep_resources(ep); 3598 return ret; 3599 } 3600 3601 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb, 3602 struct cpl_fw6_msg_ofld_connection_wr_rpl *req) 3603 { 3604 struct c4iw_ep *ep; 3605 int atid = be32_to_cpu(req->tid); 3606 3607 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids, 3608 (__force u32) req->tid); 3609 if (!ep) 3610 return; 3611 3612 switch (req->retval) { 3613 case FW_ENOMEM: 3614 set_bit(ACT_RETRY_NOMEM, &ep->com.history); 3615 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 3616 send_fw_act_open_req(ep, atid); 3617 return; 3618 } 3619 case FW_EADDRINUSE: 3620 set_bit(ACT_RETRY_INUSE, &ep->com.history); 3621 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 3622 send_fw_act_open_req(ep, atid); 3623 return; 3624 } 3625 break; 3626 default: 3627 pr_info("%s unexpected ofld conn wr retval %d\n", 3628 __func__, req->retval); 3629 break; 3630 } 3631 pr_err("active ofld_connect_wr failure %d atid %d\n", 3632 req->retval, atid); 3633 mutex_lock(&dev->rdev.stats.lock); 3634 dev->rdev.stats.act_ofld_conn_fails++; 3635 mutex_unlock(&dev->rdev.stats.lock); 3636 connect_reply_upcall(ep, status2errno(req->retval)); 3637 state_set(&ep->com, DEAD); 3638 if (ep->com.remote_addr.ss_family == AF_INET6) { 3639 struct sockaddr_in6 *sin6 = 3640 (struct sockaddr_in6 *)&ep->com.local_addr; 3641 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3642 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3643 } 3644 remove_handle(dev, &dev->atid_idr, atid); 3645 cxgb4_free_atid(dev->rdev.lldi.tids, atid); 3646 dst_release(ep->dst); 3647 cxgb4_l2t_release(ep->l2t); 3648 c4iw_put_ep(&ep->com); 3649 } 3650 3651 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb, 3652 struct cpl_fw6_msg_ofld_connection_wr_rpl *req) 3653 { 3654 struct sk_buff *rpl_skb; 3655 struct cpl_pass_accept_req *cpl; 3656 int ret; 3657 3658 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie; 3659 BUG_ON(!rpl_skb); 3660 if (req->retval) { 3661 PDBG("%s passive open failure %d\n", __func__, req->retval); 3662 mutex_lock(&dev->rdev.stats.lock); 3663 dev->rdev.stats.pas_ofld_conn_fails++; 3664 mutex_unlock(&dev->rdev.stats.lock); 3665 kfree_skb(rpl_skb); 3666 } else { 3667 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb); 3668 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 3669 (__force u32) htonl( 3670 (__force u32) req->tid))); 3671 ret = pass_accept_req(dev, rpl_skb); 3672 if (!ret) 3673 kfree_skb(rpl_skb); 3674 } 3675 return; 3676 } 3677 3678 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb) 3679 { 3680 struct cpl_fw6_msg *rpl = cplhdr(skb); 3681 struct cpl_fw6_msg_ofld_connection_wr_rpl *req; 3682 3683 switch (rpl->type) { 3684 case FW6_TYPE_CQE: 3685 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]); 3686 break; 3687 case FW6_TYPE_OFLD_CONNECTION_WR_RPL: 3688 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data; 3689 switch (req->t_state) { 3690 case TCP_SYN_SENT: 3691 active_ofld_conn_reply(dev, skb, req); 3692 break; 3693 case TCP_SYN_RECV: 3694 passive_ofld_conn_reply(dev, skb, req); 3695 break; 3696 default: 3697 pr_err("%s unexpected ofld conn wr state %d\n", 3698 __func__, req->t_state); 3699 break; 3700 } 3701 break; 3702 } 3703 return 0; 3704 } 3705 3706 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos) 3707 { 3708 __be32 l2info; 3709 __be16 hdr_len, vlantag, len; 3710 u16 eth_hdr_len; 3711 int tcp_hdr_len, ip_hdr_len; 3712 u8 intf; 3713 struct cpl_rx_pkt *cpl = cplhdr(skb); 3714 struct cpl_pass_accept_req *req; 3715 struct tcp_options_received tmp_opt; 3716 struct c4iw_dev *dev; 3717 enum chip_type type; 3718 3719 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *))); 3720 /* Store values from cpl_rx_pkt in temporary location. */ 3721 vlantag = cpl->vlan; 3722 len = cpl->len; 3723 l2info = cpl->l2info; 3724 hdr_len = cpl->hdr_len; 3725 intf = cpl->iff; 3726 3727 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header)); 3728 3729 /* 3730 * We need to parse the TCP options from SYN packet. 3731 * to generate cpl_pass_accept_req. 3732 */ 3733 memset(&tmp_opt, 0, sizeof(tmp_opt)); 3734 tcp_clear_options(&tmp_opt); 3735 tcp_parse_options(skb, &tmp_opt, 0, NULL); 3736 3737 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req)); 3738 memset(req, 0, sizeof(*req)); 3739 req->l2info = cpu_to_be16(SYN_INTF_V(intf) | 3740 SYN_MAC_IDX_V(RX_MACIDX_G( 3741 be32_to_cpu(l2info))) | 3742 SYN_XACT_MATCH_F); 3743 type = dev->rdev.lldi.adapter_type; 3744 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len)); 3745 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len)); 3746 req->hdr_len = 3747 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info)))); 3748 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) { 3749 eth_hdr_len = is_t4(type) ? 3750 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) : 3751 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info)); 3752 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) | 3753 IP_HDR_LEN_V(ip_hdr_len) | 3754 ETH_HDR_LEN_V(eth_hdr_len)); 3755 } else { /* T6 and later */ 3756 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info)); 3757 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) | 3758 T6_IP_HDR_LEN_V(ip_hdr_len) | 3759 T6_ETH_HDR_LEN_V(eth_hdr_len)); 3760 } 3761 req->vlan = vlantag; 3762 req->len = len; 3763 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) | 3764 PASS_OPEN_TOS_V(tos)); 3765 req->tcpopt.mss = htons(tmp_opt.mss_clamp); 3766 if (tmp_opt.wscale_ok) 3767 req->tcpopt.wsf = tmp_opt.snd_wscale; 3768 req->tcpopt.tstamp = tmp_opt.saw_tstamp; 3769 if (tmp_opt.sack_ok) 3770 req->tcpopt.sack = 1; 3771 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0)); 3772 return; 3773 } 3774 3775 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb, 3776 __be32 laddr, __be16 lport, 3777 __be32 raddr, __be16 rport, 3778 u32 rcv_isn, u32 filter, u16 window, 3779 u32 rss_qid, u8 port_id) 3780 { 3781 struct sk_buff *req_skb; 3782 struct fw_ofld_connection_wr *req; 3783 struct cpl_pass_accept_req *cpl = cplhdr(skb); 3784 int ret; 3785 3786 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL); 3787 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req)); 3788 memset(req, 0, sizeof(*req)); 3789 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F); 3790 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16))); 3791 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F); 3792 req->le.filter = (__force __be32) filter; 3793 req->le.lport = lport; 3794 req->le.pport = rport; 3795 req->le.u.ipv4.lip = laddr; 3796 req->le.u.ipv4.pip = raddr; 3797 req->tcb.rcv_nxt = htonl(rcv_isn + 1); 3798 req->tcb.rcv_adv = htons(window); 3799 req->tcb.t_state_to_astid = 3800 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) | 3801 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) | 3802 FW_OFLD_CONNECTION_WR_ASTID_V( 3803 PASS_OPEN_TID_G(ntohl(cpl->tos_stid)))); 3804 3805 /* 3806 * We store the qid in opt2 which will be used by the firmware 3807 * to send us the wr response. 3808 */ 3809 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid)); 3810 3811 /* 3812 * We initialize the MSS index in TCB to 0xF. 3813 * So that when driver sends cpl_pass_accept_rpl 3814 * TCB picks up the correct value. If this was 0 3815 * TP will ignore any value > 0 for MSS index. 3816 */ 3817 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF)); 3818 req->cookie = (uintptr_t)skb; 3819 3820 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id); 3821 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb); 3822 if (ret < 0) { 3823 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__, 3824 ret); 3825 kfree_skb(skb); 3826 kfree_skb(req_skb); 3827 } 3828 } 3829 3830 /* 3831 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt 3832 * messages when a filter is being used instead of server to 3833 * redirect a syn packet. When packets hit filter they are redirected 3834 * to the offload queue and driver tries to establish the connection 3835 * using firmware work request. 3836 */ 3837 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb) 3838 { 3839 int stid; 3840 unsigned int filter; 3841 struct ethhdr *eh = NULL; 3842 struct vlan_ethhdr *vlan_eh = NULL; 3843 struct iphdr *iph; 3844 struct tcphdr *tcph; 3845 struct rss_header *rss = (void *)skb->data; 3846 struct cpl_rx_pkt *cpl = (void *)skb->data; 3847 struct cpl_pass_accept_req *req = (void *)(rss + 1); 3848 struct l2t_entry *e; 3849 struct dst_entry *dst; 3850 struct c4iw_ep *lep = NULL; 3851 u16 window; 3852 struct port_info *pi; 3853 struct net_device *pdev; 3854 u16 rss_qid, eth_hdr_len; 3855 int step; 3856 u32 tx_chan; 3857 struct neighbour *neigh; 3858 3859 /* Drop all non-SYN packets */ 3860 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F))) 3861 goto reject; 3862 3863 /* 3864 * Drop all packets which did not hit the filter. 3865 * Unlikely to happen. 3866 */ 3867 if (!(rss->filter_hit && rss->filter_tid)) 3868 goto reject; 3869 3870 /* 3871 * Calculate the server tid from filter hit index from cpl_rx_pkt. 3872 */ 3873 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val); 3874 3875 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid); 3876 if (!lep) { 3877 PDBG("%s connect request on invalid stid %d\n", __func__, stid); 3878 goto reject; 3879 } 3880 3881 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) { 3882 case CHELSIO_T4: 3883 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info)); 3884 break; 3885 case CHELSIO_T5: 3886 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info)); 3887 break; 3888 case CHELSIO_T6: 3889 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info)); 3890 break; 3891 default: 3892 pr_err("T%d Chip is not supported\n", 3893 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)); 3894 goto reject; 3895 } 3896 3897 if (eth_hdr_len == ETH_HLEN) { 3898 eh = (struct ethhdr *)(req + 1); 3899 iph = (struct iphdr *)(eh + 1); 3900 } else { 3901 vlan_eh = (struct vlan_ethhdr *)(req + 1); 3902 iph = (struct iphdr *)(vlan_eh + 1); 3903 skb->vlan_tci = ntohs(cpl->vlan); 3904 } 3905 3906 if (iph->version != 0x4) 3907 goto reject; 3908 3909 tcph = (struct tcphdr *)(iph + 1); 3910 skb_set_network_header(skb, (void *)iph - (void *)rss); 3911 skb_set_transport_header(skb, (void *)tcph - (void *)rss); 3912 skb_get(skb); 3913 3914 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__, 3915 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr), 3916 ntohs(tcph->source), iph->tos); 3917 3918 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev, 3919 iph->daddr, iph->saddr, tcph->dest, 3920 tcph->source, iph->tos); 3921 if (!dst) { 3922 pr_err("%s - failed to find dst entry!\n", 3923 __func__); 3924 goto reject; 3925 } 3926 neigh = dst_neigh_lookup_skb(dst, skb); 3927 3928 if (!neigh) { 3929 pr_err("%s - failed to allocate neigh!\n", 3930 __func__); 3931 goto free_dst; 3932 } 3933 3934 if (neigh->dev->flags & IFF_LOOPBACK) { 3935 pdev = ip_dev_find(&init_net, iph->daddr); 3936 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, 3937 pdev, 0); 3938 pi = (struct port_info *)netdev_priv(pdev); 3939 tx_chan = cxgb4_port_chan(pdev); 3940 dev_put(pdev); 3941 } else { 3942 pdev = get_real_dev(neigh->dev); 3943 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, 3944 pdev, 0); 3945 pi = (struct port_info *)netdev_priv(pdev); 3946 tx_chan = cxgb4_port_chan(pdev); 3947 } 3948 neigh_release(neigh); 3949 if (!e) { 3950 pr_err("%s - failed to allocate l2t entry!\n", 3951 __func__); 3952 goto free_dst; 3953 } 3954 3955 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan; 3956 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step]; 3957 window = (__force u16) htons((__force u16)tcph->window); 3958 3959 /* Calcuate filter portion for LE region. */ 3960 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple( 3961 dev->rdev.lldi.ports[0], 3962 e)); 3963 3964 /* 3965 * Synthesize the cpl_pass_accept_req. We have everything except the 3966 * TID. Once firmware sends a reply with TID we update the TID field 3967 * in cpl and pass it through the regular cpl_pass_accept_req path. 3968 */ 3969 build_cpl_pass_accept_req(skb, stid, iph->tos); 3970 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr, 3971 tcph->source, ntohl(tcph->seq), filter, window, 3972 rss_qid, pi->port_id); 3973 cxgb4_l2t_release(e); 3974 free_dst: 3975 dst_release(dst); 3976 reject: 3977 if (lep) 3978 c4iw_put_ep(&lep->com); 3979 return 0; 3980 } 3981 3982 /* 3983 * These are the real handlers that are called from a 3984 * work queue. 3985 */ 3986 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = { 3987 [CPL_ACT_ESTABLISH] = act_establish, 3988 [CPL_ACT_OPEN_RPL] = act_open_rpl, 3989 [CPL_RX_DATA] = rx_data, 3990 [CPL_ABORT_RPL_RSS] = abort_rpl, 3991 [CPL_ABORT_RPL] = abort_rpl, 3992 [CPL_PASS_OPEN_RPL] = pass_open_rpl, 3993 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl, 3994 [CPL_PASS_ACCEPT_REQ] = pass_accept_req, 3995 [CPL_PASS_ESTABLISH] = pass_establish, 3996 [CPL_PEER_CLOSE] = peer_close, 3997 [CPL_ABORT_REQ_RSS] = peer_abort, 3998 [CPL_CLOSE_CON_RPL] = close_con_rpl, 3999 [CPL_RDMA_TERMINATE] = terminate, 4000 [CPL_FW4_ACK] = fw4_ack, 4001 [CPL_FW6_MSG] = deferred_fw6_msg, 4002 [CPL_RX_PKT] = rx_pkt, 4003 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe, 4004 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe 4005 }; 4006 4007 static void process_timeout(struct c4iw_ep *ep) 4008 { 4009 struct c4iw_qp_attributes attrs; 4010 int abort = 1; 4011 4012 mutex_lock(&ep->com.mutex); 4013 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid, 4014 ep->com.state); 4015 set_bit(TIMEDOUT, &ep->com.history); 4016 switch (ep->com.state) { 4017 case MPA_REQ_SENT: 4018 connect_reply_upcall(ep, -ETIMEDOUT); 4019 break; 4020 case MPA_REQ_WAIT: 4021 case MPA_REQ_RCVD: 4022 case MPA_REP_SENT: 4023 case FPDU_MODE: 4024 break; 4025 case CLOSING: 4026 case MORIBUND: 4027 if (ep->com.cm_id && ep->com.qp) { 4028 attrs.next_state = C4IW_QP_STATE_ERROR; 4029 c4iw_modify_qp(ep->com.qp->rhp, 4030 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 4031 &attrs, 1); 4032 } 4033 close_complete_upcall(ep, -ETIMEDOUT); 4034 break; 4035 case ABORTING: 4036 case DEAD: 4037 4038 /* 4039 * These states are expected if the ep timed out at the same 4040 * time as another thread was calling stop_ep_timer(). 4041 * So we silently do nothing for these states. 4042 */ 4043 abort = 0; 4044 break; 4045 default: 4046 WARN(1, "%s unexpected state ep %p tid %u state %u\n", 4047 __func__, ep, ep->hwtid, ep->com.state); 4048 abort = 0; 4049 } 4050 mutex_unlock(&ep->com.mutex); 4051 if (abort) 4052 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 4053 c4iw_put_ep(&ep->com); 4054 } 4055 4056 static void process_timedout_eps(void) 4057 { 4058 struct c4iw_ep *ep; 4059 4060 spin_lock_irq(&timeout_lock); 4061 while (!list_empty(&timeout_list)) { 4062 struct list_head *tmp; 4063 4064 tmp = timeout_list.next; 4065 list_del(tmp); 4066 tmp->next = NULL; 4067 tmp->prev = NULL; 4068 spin_unlock_irq(&timeout_lock); 4069 ep = list_entry(tmp, struct c4iw_ep, entry); 4070 process_timeout(ep); 4071 spin_lock_irq(&timeout_lock); 4072 } 4073 spin_unlock_irq(&timeout_lock); 4074 } 4075 4076 static void process_work(struct work_struct *work) 4077 { 4078 struct sk_buff *skb = NULL; 4079 struct c4iw_dev *dev; 4080 struct cpl_act_establish *rpl; 4081 unsigned int opcode; 4082 int ret; 4083 4084 process_timedout_eps(); 4085 while ((skb = skb_dequeue(&rxq))) { 4086 rpl = cplhdr(skb); 4087 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *))); 4088 opcode = rpl->ot.opcode; 4089 4090 BUG_ON(!work_handlers[opcode]); 4091 ret = work_handlers[opcode](dev, skb); 4092 if (!ret) 4093 kfree_skb(skb); 4094 process_timedout_eps(); 4095 } 4096 } 4097 4098 static DECLARE_WORK(skb_work, process_work); 4099 4100 static void ep_timeout(unsigned long arg) 4101 { 4102 struct c4iw_ep *ep = (struct c4iw_ep *)arg; 4103 int kickit = 0; 4104 4105 spin_lock(&timeout_lock); 4106 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 4107 /* 4108 * Only insert if it is not already on the list. 4109 */ 4110 if (!ep->entry.next) { 4111 list_add_tail(&ep->entry, &timeout_list); 4112 kickit = 1; 4113 } 4114 } 4115 spin_unlock(&timeout_lock); 4116 if (kickit) 4117 queue_work(workq, &skb_work); 4118 } 4119 4120 /* 4121 * All the CM events are handled on a work queue to have a safe context. 4122 */ 4123 static int sched(struct c4iw_dev *dev, struct sk_buff *skb) 4124 { 4125 4126 /* 4127 * Save dev in the skb->cb area. 4128 */ 4129 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev; 4130 4131 /* 4132 * Queue the skb and schedule the worker thread. 4133 */ 4134 skb_queue_tail(&rxq, skb); 4135 queue_work(workq, &skb_work); 4136 return 0; 4137 } 4138 4139 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 4140 { 4141 struct cpl_set_tcb_rpl *rpl = cplhdr(skb); 4142 4143 if (rpl->status != CPL_ERR_NONE) { 4144 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u " 4145 "for tid %u\n", rpl->status, GET_TID(rpl)); 4146 } 4147 kfree_skb(skb); 4148 return 0; 4149 } 4150 4151 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb) 4152 { 4153 struct cpl_fw6_msg *rpl = cplhdr(skb); 4154 struct c4iw_wr_wait *wr_waitp; 4155 int ret; 4156 4157 PDBG("%s type %u\n", __func__, rpl->type); 4158 4159 switch (rpl->type) { 4160 case FW6_TYPE_WR_RPL: 4161 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff); 4162 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1]; 4163 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret); 4164 if (wr_waitp) 4165 c4iw_wake_up(wr_waitp, ret ? -ret : 0); 4166 kfree_skb(skb); 4167 break; 4168 case FW6_TYPE_CQE: 4169 case FW6_TYPE_OFLD_CONNECTION_WR_RPL: 4170 sched(dev, skb); 4171 break; 4172 default: 4173 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__, 4174 rpl->type); 4175 kfree_skb(skb); 4176 break; 4177 } 4178 return 0; 4179 } 4180 4181 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb) 4182 { 4183 struct cpl_abort_req_rss *req = cplhdr(skb); 4184 struct c4iw_ep *ep; 4185 unsigned int tid = GET_TID(req); 4186 4187 ep = get_ep_from_tid(dev, tid); 4188 /* This EP will be dereferenced in peer_abort() */ 4189 if (!ep) { 4190 printk(KERN_WARNING MOD 4191 "Abort on non-existent endpoint, tid %d\n", tid); 4192 kfree_skb(skb); 4193 return 0; 4194 } 4195 if (cxgb_is_neg_adv(req->status)) { 4196 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n", 4197 __func__, ep->hwtid, req->status, 4198 neg_adv_str(req->status)); 4199 goto out; 4200 } 4201 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid, 4202 ep->com.state); 4203 4204 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 4205 out: 4206 sched(dev, skb); 4207 return 0; 4208 } 4209 4210 /* 4211 * Most upcalls from the T4 Core go to sched() to 4212 * schedule the processing on a work queue. 4213 */ 4214 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = { 4215 [CPL_ACT_ESTABLISH] = sched, 4216 [CPL_ACT_OPEN_RPL] = sched, 4217 [CPL_RX_DATA] = sched, 4218 [CPL_ABORT_RPL_RSS] = sched, 4219 [CPL_ABORT_RPL] = sched, 4220 [CPL_PASS_OPEN_RPL] = sched, 4221 [CPL_CLOSE_LISTSRV_RPL] = sched, 4222 [CPL_PASS_ACCEPT_REQ] = sched, 4223 [CPL_PASS_ESTABLISH] = sched, 4224 [CPL_PEER_CLOSE] = sched, 4225 [CPL_CLOSE_CON_RPL] = sched, 4226 [CPL_ABORT_REQ_RSS] = peer_abort_intr, 4227 [CPL_RDMA_TERMINATE] = sched, 4228 [CPL_FW4_ACK] = sched, 4229 [CPL_SET_TCB_RPL] = set_tcb_rpl, 4230 [CPL_FW6_MSG] = fw6_msg, 4231 [CPL_RX_PKT] = sched 4232 }; 4233 4234 int __init c4iw_cm_init(void) 4235 { 4236 spin_lock_init(&timeout_lock); 4237 skb_queue_head_init(&rxq); 4238 4239 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM); 4240 if (!workq) 4241 return -ENOMEM; 4242 4243 return 0; 4244 } 4245 4246 void c4iw_cm_term(void) 4247 { 4248 WARN_ON(!list_empty(&timeout_list)); 4249 flush_workqueue(workq); 4250 destroy_workqueue(workq); 4251 } 4252